add whl file

Add support for the new NeMo Canary ASR model across multiple language bindings by introducing a Canary model configuration and setter method on the offline recognizer.

- Define Canary model config in Pascal, Go, C#, Dart and update converter functions
- Add SetConfig API for offline recognizer (Pascal, Go, C#, Dart)
- Extend CI/workflows and example scripts to test non-streaming Canary decoding

.github/scripts/test-dot-net.sh

@@ -6,6 +6,11 @@ cd ./version-test
 ./run.sh
 ls -lh
 
+cd ../non-streaming-canary-decode-files
+./run.sh
+ls -lh
+rm -rf sherpa-onnx-nemo-*
+
 cd ../offline-decode-files
 
 ./run-zipformer-ctc.sh

.github/scripts/test-nodejs-addon-npm.sh

@@ -10,6 +10,16 @@ arch=$(node -p "require('os').arch()")
 platform=$(node -p "require('os').platform()")
 node_version=$(node -p "process.versions.node.split('.')[0]")
 
+echo "----------non-streaming ASR NeMo Canary----------"
+
+curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+
+node ./test_asr_non_streaming_nemo_canary.js
+
+rm -rf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8
+
 echo "----------non-streaming ASR Zipformer CTC----------"
 curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-zipformer-ctc-zh-int8-2025-07-03.tar.bz2
 

.github/scripts/test-nodejs-npm.sh

@@ -9,6 +9,14 @@ git status
 ls -lh
 ls -lh node_modules
 
+# asr with offline nemo canary
+curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+
+node ./test-offline-nemo-canary.js
+rm -rf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8
+
 # asr with offline zipformer ctc
 curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-zipformer-ctc-zh-int8-2025-07-03.tar.bz2
 

.github/scripts/test-offline-transducer.sh

@@ -281,7 +281,39 @@ time $EXE \
   $repo/test_wavs/1.wav \
   $repo/test_wavs/8k.wav
 
-rm -rf $repo
+lm_repo_url=https://huggingface.co/ezerhouni/icefall-librispeech-rnn-lm
+log "Download pre-trained RNN-LM model from ${lm_repo_url}"
+GIT_LFS_SKIP_SMUDGE=1 git clone $lm_repo_url
+lm_repo=$(basename $lm_repo_url)
+pushd $lm_repo
+git lfs pull --include "exp/no-state-epoch-99-avg-1.onnx"
+popd
+
+bigram_repo_url=https://huggingface.co/vsd-vector/librispeech_bigram_sherpa-onnx-zipformer-large-en-2023-06-26
+log "Download bi-gram LM from ${bigram_repo_url}"
+GIT_LFS_SKIP_SMUDGE=1 git clone $bigram_repo_url
+bigramlm_repo=$(basename $bigram_repo_url)
+pushd $bigramlm_repo
+git lfs pull --include "2gram.fst"
+popd
+
+log "Start testing with LM and bi-gram LODR"
+# TODO: find test examples that change with the LODR
+time $EXE \
+  --tokens=$repo/tokens.txt \
+  --encoder=$repo/encoder-epoch-99-avg-1.onnx \
+  --decoder=$repo/decoder-epoch-99-avg-1.onnx \
+  --joiner=$repo/joiner-epoch-99-avg-1.onnx \
+  --num-threads=2 \
+  --decoding_method="modified_beam_search" \
+  --lm=$lm_repo/exp/no-state-epoch-99-avg-1.onnx \
+  --lodr-fst=$bigramlm_repo/2gram.fst \
+  --lodr-scale=-0.5  \
+  $repo/test_wavs/0.wav \
+  $repo/test_wavs/1.wav \
+  $repo/test_wavs/8k.wav
+
+rm -rf $repo $lm_repo $bigramlm_repo
 
 log "------------------------------------------------------------"
 log "Run Paraformer (Chinese)"

.github/scripts/test-online-ctc.sh

@@ -77,16 +77,6 @@ time $EXE \
   $repo/test_wavs/DEV_T0000000001.wav \
   $repo/test_wavs/DEV_T0000000002.wav
 
-log "test int8"
-
-time $EXE \
-  --debug=1 \
-  --zipformer2-ctc-model=$repo/ctc-epoch-20-avg-1-chunk-16-left-128.int8.onnx \
-  --tokens=$repo/tokens.txt \
-  $repo/test_wavs/DEV_T0000000000.wav \
-  $repo/test_wavs/DEV_T0000000001.wav \
-  $repo/test_wavs/DEV_T0000000002.wav
-
 rm -rf $repo
 
 log "------------------------------------------------------------"

.github/scripts/test-online-transducer.sh

@@ -174,7 +174,60 @@ for wave in ${waves[@]}; do
   $wave
 done
 
-rm -rf $repo
+lm_repo_url=https://huggingface.co/vsd-vector/icefall-librispeech-rnn-lm
+log "Download pre-trained RNN-LM model from ${lm_repo_url}"
+GIT_LFS_SKIP_SMUDGE=1 git clone $lm_repo_url
+lm_repo=$(basename $lm_repo_url)
+pushd $lm_repo
+git lfs pull --include "with-state-epoch-99-avg-1.onnx"
+popd
+
+bigram_repo_url=https://huggingface.co/vsd-vector/librispeech_bigram_sherpa-onnx-zipformer-large-en-2023-06-26
+log "Download bi-gram LM from ${bigram_repo_url}"
+GIT_LFS_SKIP_SMUDGE=1 git clone $bigram_repo_url
+bigramlm_repo=$(basename $bigram_repo_url)
+pushd $bigramlm_repo
+git lfs pull --include "2gram.fst"
+popd
+
+log "Start testing LODR"
+
+waves=(
+$repo/test_wavs/0.wav
+$repo/test_wavs/1.wav
+$repo/test_wavs/8k.wav
+)
+
+for wave in ${waves[@]}; do
+  time $EXE \
+  --tokens=$repo/tokens.txt \
+  --encoder=$repo/encoder-epoch-99-avg-1.onnx \
+  --decoder=$repo/decoder-epoch-99-avg-1.onnx \
+  --joiner=$repo/joiner-epoch-99-avg-1.onnx \
+  --num-threads=2 \
+  --decoding_method="modified_beam_search" \
+  --lm=$lm_repo/with-state-epoch-99-avg-1.onnx \
+  --lodr-fst=$bigramlm_repo/2gram.fst \
+  --lodr-scale=-0.5  \
+  $wave
+done
+
+for wave in ${waves[@]}; do
+  time $EXE \
+  --tokens=$repo/tokens.txt \
+  --encoder=$repo/encoder-epoch-99-avg-1.onnx \
+  --decoder=$repo/decoder-epoch-99-avg-1.onnx \
+  --joiner=$repo/joiner-epoch-99-avg-1.onnx \
+  --num-threads=2 \
+  --decoding_method="modified_beam_search" \
+  --lm=$lm_repo/with-state-epoch-99-avg-1.onnx \
+  --lodr-fst=$bigramlm_repo/2gram.fst \
+  --lodr-scale=-0.5  \
+  --lm-shallow-fusion=true \
+  $wave
+done
+
+rm -rf $repo $bigramlm_repo $lm_repo
 
 log "------------------------------------------------------------"
 log "Run streaming Zipformer transducer (Bilingual, Chinese + English)"

.github/scripts/test-python.sh

@@ -562,9 +562,39 @@ python3 ./python-api-examples/offline-decode-files.py \
   $repo/test_wavs/1.wav \
   $repo/test_wavs/8k.wav
 
+lm_repo_url=https://huggingface.co/ezerhouni/icefall-librispeech-rnn-lm
+log "Download pre-trained RNN-LM model from ${lm_repo_url}"
+GIT_LFS_SKIP_SMUDGE=1 git clone $lm_repo_url
+lm_repo=$(basename $lm_repo_url)
+pushd $lm_repo
+git lfs pull --include "exp/no-state-epoch-99-avg-1.onnx"
+popd
+
+bigram_repo_url=https://huggingface.co/vsd-vector/librispeech_bigram_sherpa-onnx-zipformer-large-en-2023-06-26
+log "Download bi-gram LM from ${bigram_repo_url}"
+GIT_LFS_SKIP_SMUDGE=1 git clone $bigram_repo_url
+bigramlm_repo=$(basename $bigram_repo_url)
+pushd $bigramlm_repo
+git lfs pull --include "2gram.fst"
+popd
+
+log "Perform offline decoding with RNN-LM and LODR"
+python3 ./python-api-examples/offline-decode-files.py \
+  --tokens=$repo/tokens.txt \
+  --encoder=$repo/encoder-epoch-99-avg-1.onnx \
+  --decoder=$repo/decoder-epoch-99-avg-1.onnx \
+  --joiner=$repo/joiner-epoch-99-avg-1.onnx \
+  --decoding-method=modified_beam_search \
+  --lm=$lm_repo/exp/no-state-epoch-99-avg-1.onnx \
+  --lodr-fst=$bigramlm_repo/2gram.fst \
+  --lodr-scale=-0.5 \
+  $repo/test_wavs/0.wav \
+  $repo/test_wavs/1.wav \
+  $repo/test_wavs/8k.wav
+
 python3 sherpa-onnx/python/tests/test_offline_recognizer.py --verbose
 
-rm -rf $repo
+rm -rf $repo $lm_repo $bigramlm_repo
 
 log "Test non-streaming paraformer models"
 

.github/workflows/c-api.yaml

@@ -127,6 +127,36 @@ jobs:
           rm -rf dict lexicon.txt test-hr.wav replace.fst
           rm -v $name
 
+      - name: Test NeMo Canary
+        shell: bash
+        run: |
+          name=nemo-canary-c-api
+          gcc -o $name ./c-api-examples/$name.c \
+            -I ./build/install/include \
+            -L ./build/install/lib/ \
+            -l sherpa-onnx-c-api \
+            -l onnxruntime
+
+          ls -lh $name
+
+          if [[ ${{ matrix.os }} == ubuntu-latest || ${{ matrix.os }} == ubuntu-22.04-arm ]]; then
+            ldd ./$name
+            echo "----"
+            readelf -d ./$name
+          fi
+
+          curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+          tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+          rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+
+          export LD_LIBRARY_PATH=$PWD/build/install/lib:$LD_LIBRARY_PATH
+          export DYLD_LIBRARY_PATH=$PWD/build/install/lib:$DYLD_LIBRARY_PATH
+
+          ./$name
+
+          rm $name
+          rm -rf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8
+
       - name: Test Dolphin CTC
         shell: bash
         run: |

.github/workflows/cxx-api.yaml

@@ -87,6 +87,40 @@ jobs:
             otool -L ./install/lib/libsherpa-onnx-cxx-api.dylib
           fi
 
+      - name: Test NeMo Canary
+        shell: bash
+        run: |
+          name=nemo-canary-cxx-api
+          g++ -std=c++17 -o $name ./cxx-api-examples/$name.cc \
+            -I ./build/install/include \
+            -L ./build/install/lib/ \
+            -l sherpa-onnx-cxx-api \
+            -l sherpa-onnx-c-api \
+            -l onnxruntime
+
+          ls -lh $name
+
+          if [[ ${{ matrix.os }} == ubuntu-latest || ${{ matrix.os }} == ubuntu-22.04-arm ]]; then
+            ldd ./$name
+            echo "----"
+            readelf -d ./$name
+          fi
+
+          curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+          tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+          rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+
+          ls -lh sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8
+          echo "---"
+
+          export LD_LIBRARY_PATH=$PWD/build/install/lib:$LD_LIBRARY_PATH
+          export DYLD_LIBRARY_PATH=$PWD/build/install/lib:$DYLD_LIBRARY_PATH
+
+          ./$name
+
+          rm -rf sherpa-onnx-nemo-canary-*
+          rm -v ./$name
+
       - name: Test streaming zipformer with Homophone replacer
         shell: bash
         run: |

.github/workflows/export-nemo-fast-conformer-hybrid-transducer-ctc-non-streaming.yaml

@@ -61,6 +61,11 @@ jobs:
               sherpa-onnx-nemo-fast-conformer-ctc-en-de-es-fr-14288
               sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k
               sherpa-onnx-nemo-parakeet_tdt_ctc_110m-en-36000
+              sherpa-onnx-nemo-fast-conformer-ctc-en-24500-int8
+              sherpa-onnx-nemo-fast-conformer-ctc-es-1424-int8
+              sherpa-onnx-nemo-fast-conformer-ctc-en-de-es-fr-14288-int8
+              sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k-int8
+              sherpa-onnx-nemo-parakeet_tdt_ctc_110m-en-36000-int8
             )
 
             for m in ${models[@]}; do
@@ -89,6 +94,11 @@ jobs:
             sherpa-onnx-nemo-fast-conformer-ctc-en-de-es-fr-14288
             sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k
             sherpa-onnx-nemo-parakeet_tdt_ctc_110m-en-36000
+            sherpa-onnx-nemo-fast-conformer-ctc-en-24500-int8
+            sherpa-onnx-nemo-fast-conformer-ctc-es-1424-int8
+            sherpa-onnx-nemo-fast-conformer-ctc-en-de-es-fr-14288-int8
+            sherpa-onnx-nemo-fast-conformer-ctc-be-de-en-es-fr-hr-it-pl-ru-uk-20k-int8
+            sherpa-onnx-nemo-parakeet_tdt_ctc_110m-en-36000-int8
           )
           for d in ${dirs[@]}; do
             tar cjvf ${d}.tar.bz2 ./$d

.github/workflows/export-nemo-fast-conformer-hybrid-transducer-ctc.yaml

@@ -54,13 +54,18 @@ jobs:
           curl -SL -O https://hf-mirror.com/csukuangfj/sherpa-onnx-nemo-ctc-en-conformer-small/resolve/main/test_wavs/trans.txt
           popd
 
-          cp -av test_wavs ./sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-80ms
+          names=(
-          cp -av test_wavs ./sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-480ms
+            sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-80ms
-          cp -av test_wavs ./sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-1040ms
+            sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-480ms
-
+            sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-1040ms
-          tar cjvf sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-80ms.tar.bz2 sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-80ms
+            sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-80ms-int8
-          tar cjvf sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-480ms.tar.bz2 sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-480ms
+            sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-480ms-int8
-          tar cjvf sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-1040ms.tar.bz2 sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-1040ms
+            sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-1040ms-int8
+          )
+          for d in ${names[@]}; do
+            cp -av test_wavs $d/
+            tar cjvf $d.tar.bz2 $d
+          done
 
       - name: Release
         uses: svenstaro/upload-release-action@v2
@@ -71,3 +76,41 @@ jobs:
           repo_name: k2-fsa/sherpa-onnx
           repo_token: ${{ secrets.UPLOAD_GH_SHERPA_ONNX_TOKEN }}
           tag: asr-models
+
+      - name: Publish to huggingface
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        uses: nick-fields/retry@v3
+        with:
+          max_attempts: 20
+          timeout_seconds: 200
+          shell: bash
+          command: |
+            git config --global user.email "csukuangfj@gmail.com"
+            git config --global user.name "Fangjun Kuang"
+
+            models=(
+              sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-80ms
+              sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-480ms
+              sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-1040ms
+              sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-80ms-int8
+              sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-480ms-int8
+              sherpa-onnx-nemo-streaming-fast-conformer-ctc-en-1040ms-int8
+            )
+
+            for m in ${models[@]}; do
+              rm -rf huggingface
+              export GIT_LFS_SKIP_SMUDGE=1
+              export GIT_CLONE_PROTECTION_ACTIVE=false
+              git clone https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$m huggingface
+              cp -av $m/* huggingface
+              cd huggingface
+              git lfs track "*.onnx"
+              git lfs track "*.wav"
+              git status
+              git add .
+              git status
+              git commit -m "first commit"
+              git push https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$m main
+              cd ..
+            done

.github/workflows/export-nemo-fast-conformer-hybrid-transducer-transducer-non-streaming.yaml

@@ -61,6 +61,11 @@ jobs:
               sherpa-onnx-nemo-fast-conformer-transducer-en-de-es-fr-14288
               sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k
               sherpa-onnx-nemo-parakeet_tdt_transducer_110m-en-36000
+              sherpa-onnx-nemo-fast-conformer-transducer-en-24500-int8
+              sherpa-onnx-nemo-fast-conformer-transducer-es-1424-int8
+              sherpa-onnx-nemo-fast-conformer-transducer-en-de-es-fr-14288-int8
+              sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k-int8
+              sherpa-onnx-nemo-parakeet_tdt_transducer_110m-en-36000-int8
             )
 
             for m in ${models[@]}; do
@@ -88,6 +93,11 @@ jobs:
             sherpa-onnx-nemo-fast-conformer-transducer-en-de-es-fr-14288
             sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k
             sherpa-onnx-nemo-parakeet_tdt_transducer_110m-en-36000
+            sherpa-onnx-nemo-fast-conformer-transducer-en-24500-int8
+            sherpa-onnx-nemo-fast-conformer-transducer-es-1424-int8
+            sherpa-onnx-nemo-fast-conformer-transducer-en-de-es-fr-14288-int8
+            sherpa-onnx-nemo-fast-conformer-transducer-be-de-en-es-fr-hr-it-pl-ru-uk-20k-int8
+            sherpa-onnx-nemo-parakeet_tdt_transducer_110m-en-36000-int8
           )
           for d in ${dirs[@]}; do
             tar cjvf ${d}.tar.bz2 ./$d

.github/workflows/export-nemo-fast-conformer-hybrid-transducer-transducer.yaml

@@ -54,13 +54,18 @@ jobs:
           curl -SL -O https://hf-mirror.com/csukuangfj/sherpa-onnx-nemo-ctc-en-conformer-small/resolve/main/test_wavs/trans.txt
           popd
 
-          cp -av test_wavs ./sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-80ms
+          models=(
-          cp -av test_wavs ./sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-480ms
+            sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-80ms
-          cp -av test_wavs ./sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-1040ms
+            sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-480ms
-
+            sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-1040ms
-          tar cjvf sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-80ms.tar.bz2 sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-80ms
+            sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-80ms-int8
-          tar cjvf sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-480ms.tar.bz2 sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-480ms
+            sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-480ms-int8
-          tar cjvf sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-1040ms.tar.bz2 sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-1040ms
+            sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-1040ms-int8
+          )
+          for m in ${models[@]}; do
+            cp -av test_wavs $m
+            tar cjvf $m.tar.bz2 $m
+          done
 
       - name: Release
         uses: svenstaro/upload-release-action@v2
@@ -71,3 +76,41 @@ jobs:
           repo_name: k2-fsa/sherpa-onnx
           repo_token: ${{ secrets.UPLOAD_GH_SHERPA_ONNX_TOKEN }}
           tag: asr-models
+
+      - name: Publish to huggingface
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        uses: nick-fields/retry@v3
+        with:
+          max_attempts: 20
+          timeout_seconds: 200
+          shell: bash
+          command: |
+            git config --global user.email "csukuangfj@gmail.com"
+            git config --global user.name "Fangjun Kuang"
+
+            models=(
+              sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-80ms
+              sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-480ms
+              sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-1040ms
+              sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-80ms-int8
+              sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-480ms-int8
+              sherpa-onnx-nemo-streaming-fast-conformer-transducer-en-1040ms-int8
+            )
+
+            for m in ${models[@]}; do
+              rm -rf huggingface
+              export GIT_LFS_SKIP_SMUDGE=1
+              export GIT_CLONE_PROTECTION_ACTIVE=false
+              git clone https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$m huggingface
+              cp -av $m/* huggingface
+              cd huggingface
+              git lfs track "*.onnx"
+              git lfs track "*.wav"
+              git status
+              git add .
+              git status
+              git commit -m "first commit"
+              git push https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$m main
+              cd ..
+            done

.github/workflows/export-nemo-parakeet-tdt.yaml

@@ -0,0 +1,105 @@
+name: export-nemo-parakeet-tdt
+
+on:
+  push:
+    branches:
+      - refactor-export-nemo
+  workflow_dispatch:
+
+concurrency:
+  group: export-nemo-parakeet-tdt-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  export-nemo-parakeet-tdt-0_6b-v2:
+    if: github.repository_owner == 'k2-fsa' || github.repository_owner == 'csukuangfj'
+    name: parakeet tdt
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [macos-latest]
+        python-version: ["3.10"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Setup Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install python dependencies
+        shell: bash
+        run: |
+          pip install \
+            nemo_toolkit['asr'] \
+            "numpy<2" \
+            ipython \
+            kaldi-native-fbank \
+            librosa \
+            onnx==1.17.0 \
+            onnxmltools==1.13.0 \
+            onnxruntime==1.17.1 \
+            soundfile
+
+      - name: Run
+        shell: bash
+        run: |
+          cd scripts/nemo/parakeet-tdt_ctc-0.6b-ja
+          ./run-ctc.sh
+
+      - name: Collect files
+        shell: bash
+        run: |
+          models=(
+            sherpa-onnx-nemo-parakeet-tdt_ctc-0.6b-ja-35000-int8
+          )
+          for m in ${models[@]}; do
+            mv -v scripts/nemo/parakeet-tdt_ctc-0.6b-ja/$m .
+            tar cjfv $m.tar.bz2 $m
+          done
+
+
+      - name: Publish to huggingface
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        uses: nick-fields/retry@v3
+        with:
+          max_attempts: 20
+          timeout_seconds: 200
+          shell: bash
+          command: |
+            git config --global user.email "csukuangfj@gmail.com"
+            git config --global user.name "Fangjun Kuang"
+
+            models=(
+              sherpa-onnx-nemo-parakeet-tdt_ctc-0.6b-ja-35000-int8
+            )
+
+            for m in ${models[@]}; do
+              rm -rf huggingface
+              export GIT_LFS_SKIP_SMUDGE=1
+              export GIT_CLONE_PROTECTION_ACTIVE=false
+              git clone https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$m huggingface
+              cp -av $m/* huggingface
+              cd huggingface
+              git lfs track "*.onnx"
+              git lfs track "*.wav"
+              git status
+              git add .
+              git status
+              git commit -m "first commit"
+              git push https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$m main
+              cd ..
+            done
+
+      - name: Release
+        uses: svenstaro/upload-release-action@v2
+        with:
+          file_glob: true
+          file: ./*.tar.bz2
+          overwrite: true
+          repo_name: k2-fsa/sherpa-onnx
+          repo_token: ${{ secrets.UPLOAD_GH_SHERPA_ONNX_TOKEN }}
+          tag: asr-models

.github/workflows/pascal.yaml

@@ -156,6 +156,10 @@ jobs:
 
           pushd non-streaming-asr
 
+          ./run-nemo-canary.sh
+          rm -rf sherpa-onnx-*
+          echo "---"
+
           ./run-zipformer-ctc.sh
           rm -rf sherpa-onnx-*
           echo "---"

.github/workflows/run-java-test.yaml

@@ -117,6 +117,13 @@ jobs:
           cd ./java-api-examples
           ./run-version-test.sh
 
+      - name:  Run java test (Nemo Canary)
+        shell: bash
+        run: |
+          cd ./java-api-examples
+          ./run-non-streaming-decode-file-nemo-canary.sh
+          rm -rf sherpa-onnx-nemo-*
+
       - name:  Run java test (Non-streaming SenseVoice with homophone replacer)
         shell: bash
         run: |

.github/workflows/test-go-package.yaml

@@ -76,6 +76,14 @@ jobs:
         run: |
           gcc --version
 
+      - name: Test NeMo Canary ASR
+        if: matrix.os != 'windows-latest'
+        shell: bash
+        run: |
+          cd go-api-examples/non-streaming-canary-decode-files
+          ./run.sh
+          rm -rf sherpa-onnx-nemo-*
+
       - name: Test speech enhancement (GTCRN)
         if: matrix.os != 'windows-latest'
         shell: bash

.github/workflows/test-go.yaml

@@ -108,6 +108,7 @@ jobs:
             cp -v ../scripts/go/_internal/lib/x86_64-pc-windows-gnu/*.dll ../scripts/go/_internal/add-punctuation
             cp -v ../scripts/go/_internal/lib/x86_64-pc-windows-gnu/*.dll ../scripts/go/_internal/audio-tagging
             cp -v ../scripts/go/_internal/lib/x86_64-pc-windows-gnu/*.dll ../scripts/go/_internal/keyword-spotting-from-file/
+            cp -v ../scripts/go/_internal/lib/x86_64-pc-windows-gnu/*.dll ../scripts/go/_internal/non-streaming-canary-decode-files/
             cp -v ../scripts/go/_internal/lib/x86_64-pc-windows-gnu/*.dll ../scripts/go/_internal/non-streaming-decode-files/
             cp -v ../scripts/go/_internal/lib/x86_64-pc-windows-gnu/*.dll ../scripts/go/_internal/non-streaming-speaker-diarization/
             cp -v ../scripts/go/_internal/lib/x86_64-pc-windows-gnu/*.dll ../scripts/go/_internal/non-streaming-tts/
@@ -148,6 +149,19 @@ jobs:
           name: ${{ matrix.os }}-libs
           path: to-upload/
 
+      - name: Test non-streaming decoding files with NeMo Canary
+        shell: bash
+        run: |
+          cd scripts/go/_internal/non-streaming-canary-decode-files/
+          ls -lh
+          go mod tidy
+          cat go.mod
+          go build
+          ls -lh
+
+          ./run.sh
+          rm -rf sherpa-onnx-nemo-*
+
       - name: Test streaming decoding files
         shell: bash
         run: |

.github/workflows/upload-models.yaml

@@ -24,7 +24,45 @@ jobs:
     steps:
       - uses: actions/checkout@v4
 
+      - name: git config
+        shell: bash
+        run: |
+          git config --global user.email "csukuangfj@gmail.com"
+          git config --global user.name "Fangjun Kuang"
+
+      - name: FireRed ASR fp16
+        shell: bash
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        run: |
+          GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/csukuangfj/sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16 hf
+
+          git lfs install
+          git clone https://www.modelscope.cn/csukuangfj/sherpa-onnx-fire-red-asr-large-zh_en-fp16-2025-02-16.git ms
+
+          d=sherpa-onnx-fire-red-asr-large-zh_en-fp16-2025-02-16
+          git clone https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$d
+          mv -v hf/test_wavs $d
+          mv -v hf/README.md $d
+          mv -v hf/tokens.txt $d
+          mv -v ms/*.onnx $d
+
+          pushd $d
+          git lfs track "*.onnx"
+          git lfs track "*.wav"
+          git status
+          git add .
+          git commit -m "add models"
+          ls -lh
+          git push https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$d main
+          popd
+
+          rm -rf $d/.git
+          rm -rf $d/.gitattributes
+          tar cjvf $d.tar.bz2 $d
+
       - name: Zipformer CTC (non-streaming)
+        if: false
         shell: bash
         run: |
           git lfs install

CHANGELOG.md

@@ -1,3 +1,22 @@
+## 1.12.5
+
+* Fix typo CMAKE_EXECUTBLE_LINKER_FLAGS -> CMAKE_EXECUTABLE_LINKER_FLAGS (#2344)
+* Fix testing dart packages (#2345)
+* fix(canary): use dynamo export, single input_ids and avoid 0/1 specialization (#2348)
+* Fix TTS for Unreal Engine (#2349)
+* Update readme to include https://github.com/mawwalker/stt-server (#2350)
+* Add meta data to NeMo canary ONNX models (#2351)
+* Update README to include https://github.com/bbeyondllove/asr_server (#2353)
+* Add C++ runtime and Python API for NeMo Canary models (#2352)
+* Add C/CXX/JavaScript API for NeMo Canary models (#2357)
+* Add Java and Kotlin API for NeMo Canary models (#2359)
+* Upload fp16 onnx model files for FireRedASR (#2360)
+* Fix nemo feature normalization in test code (#2361)
+* Refactor exporting NeMo models (#2362)
+* Add LODR support to online and offline recognizers (#2026)
+* Add CXX examples for NeMo TDT ASR. (#2363)
+* Add Pascal/Go/C#/Dart API for NeMo Canary ASR models (#2367)
+
 ## 1.12.4
 
 * Refactor release scripts. (#2323)

CMakeLists.txt

@@ -14,7 +14,7 @@ project(sherpa-onnx)
 # Remember to update
 # ./CHANGELOG.md
 # ./new-release.sh
-set(SHERPA_ONNX_VERSION "1.12.4")
+set(SHERPA_ONNX_VERSION "1.12.5")
 
 # Disable warning about
 #

README.md

@@ -1,3 +1,42 @@
+## 天数智芯 天垓100 sherpa-onnx
+
+### 编译指令
+在corex:3.2.1镜像中使用以下命令
+```
+SHERPA_ONNX_CMAKE_ARGS='-DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DSHERPA_ONNX_ENABLE_GPU=ON' python3 setup.py bdist_wheel
+```
+即可编译出适用于天垓100的sherpa-onnx whl包
+
+### pip使用方式
+```shell
+pip install sherpa_onnx-1.12.5+corex3.2.1-cp310-cp310-linux_x86_64.whl
+```
+具体的推理过程代码与sherpa-onnx python包没有任何区别
+
+### 天垓100的模型适配情况
+我们在天垓100上针对sherpa-onnx offline asr部分进行了所有大类的适配，测试方式为在CPU和天垓100加速卡上对同一段长音频进行语音识别任务，获取运行时间，1-cer指标。运行环境为：
+- CPU: Intel(R) Xeon(R) Gold 6238R CPU @ 2.20GHz 使用4线程推理
+- 天垓100：在上述CPU硬件的同时下使用一张天垓100加速卡
+
+| 模型大类 | 模型名称 |CPU运行时间(秒)|天垓100运行时间(秒)|CPU 1-cer|天垓100 1-cer| 备注                           |
+|------|---------------|-----|----|-------|-------|---------------------|
+| sense_voice |sherpa-onnx-sense-voice-zh-en-ja-ko-yue-2024-07-17| 36.8381 | 3.1053 | 0.990017 | 0.990017 |    |
+| whisper | sherpa-onnx-whisper-large-v3 | 705.8439 | 204.0064 | 0.722130 | 0.732113 | 天垓100的识别准确率比CPU略高   |
+| whisper | sherpa-onnx-whisper-base | 34.8192 | 22.2912 | 0.682196 | 0.682196 |    |
+| whisper | sherpa-onnx-whisper-small | 119.6921 | 46.7079 | 0.662230 | 0.662230 |    |
+| whisper | sherpa-onnx-whisper-tiny | 18.2531 | 15.2332 | 0.584027 | 0.584027 |    |
+| whisper | sherpa-onnx-whisper-medium-aishell | 418.8124 | 141.6660 | 0.785358 | 0.783694 |   |
+| dolphin_ctc | sherpa-onnx-dolphin-small-ctc-multi-lang-2025-04-02 | 17.6391 | 2.7845 | 0.923461 | 0.923461 |    |
+| fire_red_asr | sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16 | 409.8351 | 120.9066 | 0.960067 | 0.960067 |    |
+| zipformer_ctc | sherpa-onnx-zipformer-ctc-zh-int8-2025-07-03 | 10.5072 | 6.9715 | 0.963394 | 0.966722 | 天垓100的识别准确率比CPU略高 |
+| zipformer_ctc | sherpa-onnx-zipformer-ctc-small-zh-2025-07-16 | 4.9449 | 1.7400 | 0.958403 | 0.908486 | 天垓100在该模型与CPU相比精度有一些欠缺 |
+| paraformer | sherpa-onnx-paraformer-zh-2024-03-09 | 6.4983 | 15.9283 | 0.948419 | 0.948419 | 天垓100运行速度明显慢于4线程CPU推理 |
+| telespeech_ctc | sherpa-onnx-telespeech-ctc-zh-2024-06-04 | 24.0849 | 72.0935 | 0.948419 | 0.948419 | 天垓100运行速度明显慢于4线程CPU推理 |
+| wenet_ctc | sherpa-onnx-zh-wenet-aishell2 | 3.9872 | 3.7492 | 0.813644 | 0.813644 |    |
+| transducer | sherpa-onnx-zipformer-multi-zh-hans-2023-9-2 | 6.4223 | 2.6784 | 0.920133 | 0.765391 | 天垓100在该模型与CPU相比精度有较大损失 |
+| nemo_ctc | sherpa-onnx-nemo-parakeet_tdt_ctc_110m-en-36000 | N/A | N/A | N/A | N/A | 无法适配 |
+| nemo_canary | sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr | N/A | N/A | N/A | N/A | 无法适配 |
+| moonshine | sherpa-onnx-moonshine-base-en-int8 | N/A | N/A | N/A | N/A | 无法适配 |
 ### Supported functions
 
 |Speech recognition| [Speech synthesis][tts-url] | [Source separation][ss-url] |

android/SherpaOnnxAar/README.md

@@ -4,8 +4,8 @@
 git clone https://github.com/k2-fsa/sherpa-onnx
 cd sherpa-onnx
 
-wget https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.4/sherpa-onnx-v1.12.4-android.tar.bz2
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.5/sherpa-onnx-v1.12.5-android.tar.bz2
-tar xvf sherpa-onnx-v1.12.4-android.tar.bz2
+tar xvf sherpa-onnx-v1.12.5-android.tar.bz2
 
 cp -v jniLibs/arm64-v8a/* android/SherpaOnnxAar/sherpa_onnx/src/main/jniLibs/arm64-v8a/
 cp -v jniLibs/armeabi-v7a/* android/SherpaOnnxAar/sherpa_onnx/src/main/jniLibs/armeabi-v7a/
@@ -16,5 +16,5 @@ cd android/SherpaOnnxAar
 
 ./gradlew :sherpa_onnx:assembleRelease
 ls -lh ./sherpa_onnx/build/outputs/aar/sherpa_onnx-release.aar
-cp ./sherpa_onnx/build/outputs/aar/sherpa_onnx-release.aar ../../sherpa-onnx-1.12.4.aar
+cp ./sherpa_onnx/build/outputs/aar/sherpa_onnx-release.aar ../../sherpa-onnx-1.12.5.aar
 ```

android/SherpaOnnxJavaDemo/app/build.gradle

@@ -34,5 +34,5 @@ dependencies {
     implementation 'pub.devrel:easypermissions:3.0.0'
     implementation 'androidx.core:core-ktx:1.7.0'
     // implementation files('/Users/fangjun/open-source/sherpa-onnx/android/SherpaOnnxAar/sherpa_onnx/build/outputs/aar/sherpa_onnx-release.aar')
-    implementation 'com.github.k2-fsa:sherpa-onnx:v1.12.4'
+    implementation 'com.github.k2-fsa:sherpa-onnx:v1.12.5'
 }

build-ios-shared.sh

@@ -242,7 +242,7 @@ for d in ios-arm64_x86_64-simulator ios-arm64; do
 	<key>CFBundlePackageType</key>
 	<string>FMWK</string>
 	<key>CFBundleShortVersionString</key>
-	<string>1.12.4</string>
+	<string>1.12.5</string>
 	<key>CFBundleSupportedPlatforms</key>
 	<array>
 		<string>iPhoneOS</string>

c-api-examples/CMakeLists.txt

@@ -53,6 +53,9 @@ target_link_libraries(whisper-c-api sherpa-onnx-c-api)
 add_executable(fire-red-asr-c-api fire-red-asr-c-api.c)
 target_link_libraries(fire-red-asr-c-api sherpa-onnx-c-api)
 
+add_executable(nemo-canary-c-api nemo-canary-c-api.c)
+target_link_libraries(nemo-canary-c-api sherpa-onnx-c-api)
+
 add_executable(sense-voice-c-api sense-voice-c-api.c)
 target_link_libraries(sense-voice-c-api sherpa-onnx-c-api)
 

c-api-examples/nemo-canary-c-api.c

@@ -0,0 +1,115 @@
+// c-api-examples/nemo-canary-c-api.c
+//
+// Copyright (c)  2025  Xiaomi Corporation
+
+// We assume you have pre-downloaded the Nemo Canary model
+// from https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
+// An example is given below:
+//
+// clang-format off
+//
+// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+// tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+// rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+//
+// clang-format on
+//
+// see https://k2-fsa.github.io/sherpa/onnx/nemo/canary.html
+// for details
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "sherpa-onnx/c-api/c-api.h"
+
+int32_t main() {
+  const char *wav_filename =
+      "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/de.wav";
+  const char *encoder_filename =
+      "sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx";
+  const char *decoder_filename =
+      "sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx";
+  const char *tokens_filename =
+      "sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt";
+  const char *provider = "cpu";
+
+  const SherpaOnnxWave *wave = SherpaOnnxReadWave(wav_filename);
+  if (wave == NULL) {
+    fprintf(stderr, "Failed to read %s\n", wav_filename);
+    return -1;
+  }
+
+  // Offline model config
+  SherpaOnnxOfflineModelConfig offline_model_config;
+  memset(&offline_model_config, 0, sizeof(offline_model_config));
+
+  // set debug to 1 to view more logs
+  offline_model_config.debug = 0;
+
+  offline_model_config.num_threads = 1;
+  offline_model_config.provider = provider;
+  offline_model_config.tokens = tokens_filename;
+  offline_model_config.canary.encoder = encoder_filename;
+  offline_model_config.canary.decoder = decoder_filename;
+
+  // so it output punctuations and cases
+  offline_model_config.canary.use_pnc = 1;
+
+  offline_model_config.canary.src_lang = "de";
+
+  // since there is a German audio, you can set tgt_lang to en or de
+  offline_model_config.canary.tgt_lang = "en";
+
+  // Recognizer config
+  SherpaOnnxOfflineRecognizerConfig recognizer_config;
+  memset(&recognizer_config, 0, sizeof(recognizer_config));
+  recognizer_config.decoding_method = "greedy_search";
+  recognizer_config.model_config = offline_model_config;
+
+  const SherpaOnnxOfflineRecognizer *recognizer =
+      SherpaOnnxCreateOfflineRecognizer(&recognizer_config);
+
+  if (recognizer == NULL) {
+    fprintf(stderr, "Please check your config!\n");
+
+    SherpaOnnxFreeWave(wave);
+
+    return -1;
+  }
+
+  const SherpaOnnxOfflineStream *stream =
+      SherpaOnnxCreateOfflineStream(recognizer);
+
+  SherpaOnnxAcceptWaveformOffline(stream, wave->sample_rate, wave->samples,
+                                  wave->num_samples);
+  SherpaOnnxDecodeOfflineStream(recognizer, stream);
+  const SherpaOnnxOfflineRecognizerResult *result =
+      SherpaOnnxGetOfflineStreamResult(stream);
+
+  fprintf(stderr, "Decoded text (English): %s\n", result->text);
+
+  SherpaOnnxDestroyOfflineRecognizerResult(result);
+  SherpaOnnxDestroyOfflineStream(stream);
+
+  // now output German text
+  recognizer_config.model_config.canary.tgt_lang = "de";
+  SherpaOnnxOfflineRecognizerSetConfig(recognizer, &recognizer_config);
+
+  stream = SherpaOnnxCreateOfflineStream(recognizer);
+
+  SherpaOnnxAcceptWaveformOffline(stream, wave->sample_rate, wave->samples,
+                                  wave->num_samples);
+  SherpaOnnxDecodeOfflineStream(recognizer, stream);
+  result = SherpaOnnxGetOfflineStreamResult(stream);
+
+  fprintf(stderr, "Decoded text (German): %s\n", result->text);
+
+  SherpaOnnxDestroyOfflineRecognizerResult(result);
+  SherpaOnnxDestroyOfflineStream(stream);
+
+  SherpaOnnxDestroyOfflineRecognizer(recognizer);
+  SherpaOnnxFreeWave(wave);
+
+  return 0;
+}

c-api-examples/streaming-ctc-buffered-tokens-c-api.c

@@ -54,7 +54,7 @@ int32_t main() {
       "DEV_T0000000000.wav";
   const char *model_filename =
       "sherpa-onnx-streaming-zipformer-ctc-multi-zh-hans-2023-12-13/"
-      "ctc-epoch-20-avg-1-chunk-16-left-128.int8.onnx";
+      "ctc-epoch-20-avg-1-chunk-16-left-128.onnx";
   const char *tokens_filename =
       "sherpa-onnx-streaming-zipformer-ctc-multi-zh-hans-2023-12-13/tokens.txt";
   const char *provider = "cpu";

cmake/cmake_extension.py

@@ -180,8 +180,8 @@ class BuildExtension(build_ext):
             if make_args == "" and system_make_args == "":
                 print("for fast compilation, run:")
                 print('export SHERPA_ONNX_MAKE_ARGS="-j"; python setup.py install')
-                print('Setting make_args to "-j4"')
+                print('Setting make_args to "-j8"')
-                make_args = "-j4"
+                make_args = "-j8"
 
             if "-G Ninja" in cmake_args:
                 build_cmd = f"""

cmake/onnxruntime-linux-x86_64-gpu.cmake

@@ -19,9 +19,9 @@ if(NOT SHERPA_ONNX_ENABLE_GPU)
 endif()
 
 
-set(onnxruntime_URL  "https://github.com/csukuangfj/onnxruntime-libs/releases/download/v1.17.1/onnxruntime-linux-x64-gpu-1.17.1-patched.zip")
+set(onnxruntime_URL  "ftp://ftp.4pd.io/pub/iluvatar/bi_v100/onnxruntime-linux-x64-gpu-1.17.1-patched.zip")
-set(onnxruntime_URL2 "https://hf-mirror.com/csukuangfj/onnxruntime-libs/resolve/main/onnxruntime-linux-x64-gpu-1.17.1-patched.zip")
+set(onnxruntime_URL2 "ftp://ftp.4pd.io/pub/iluvatar/bi_v100/onnxruntime-linux-x64-gpu-1.17.1-patched.zip")
-set(onnxruntime_HASH "SHA256=1261de176e8d9d4d2019f8fa8c732c6d11494f3c6e73168ab6d2cc0903f22551")
+set(onnxruntime_HASH "SHA256=3b07e07aa756b625892f6933acd93f9606f32347c0342e55690f859975318014")
 
 # If you don't have access to the Internet,
 # please download onnxruntime to one of the following locations.
@@ -70,7 +70,7 @@ add_library(onnxruntime SHARED IMPORTED)
 
 set_target_properties(onnxruntime PROPERTIES
   IMPORTED_LOCATION ${location_onnxruntime}
-  INTERFACE_INCLUDE_DIRECTORIES "${onnxruntime_SOURCE_DIR}/include"
+  INTERFACE_INCLUDE_DIRECTORIES "${onnxruntime_SOURCE_DIR}/include/onnxruntime"
 )
 
 find_library(location_onnxruntime_cuda_lib onnxruntime_providers_cuda

cxx-api-examples/CMakeLists.txt

@@ -27,6 +27,9 @@ target_link_libraries(moonshine-cxx-api sherpa-onnx-cxx-api)
 add_executable(sense-voice-cxx-api ./sense-voice-cxx-api.cc)
 target_link_libraries(sense-voice-cxx-api sherpa-onnx-cxx-api)
 
+add_executable(nemo-canary-cxx-api ./nemo-canary-cxx-api.cc)
+target_link_libraries(nemo-canary-cxx-api sherpa-onnx-cxx-api)
+
 if(SHERPA_ONNX_ENABLE_PORTAUDIO)
   add_executable(sense-voice-simulate-streaming-microphone-cxx-api
     ./sense-voice-simulate-streaming-microphone-cxx-api.cc
@@ -46,6 +49,15 @@ if(SHERPA_ONNX_ENABLE_PORTAUDIO)
     portaudio_static
   )
 
+  add_executable(parakeet-tdt-ctc-simulate-streaming-microphone-cxx-api
+    ./parakeet-tdt-ctc-simulate-streaming-microphone-cxx-api.cc
+    ${CMAKE_CURRENT_LIST_DIR}/../sherpa-onnx/csrc/microphone.cc
+  )
+  target_link_libraries(parakeet-tdt-ctc-simulate-streaming-microphone-cxx-api
+    sherpa-onnx-cxx-api
+    portaudio_static
+  )
+
   add_executable(zipformer-ctc-simulate-streaming-microphone-cxx-api
     ./zipformer-ctc-simulate-streaming-microphone-cxx-api.cc
     ${CMAKE_CURRENT_LIST_DIR}/../sherpa-onnx/csrc/microphone.cc
@@ -54,6 +66,15 @@ if(SHERPA_ONNX_ENABLE_PORTAUDIO)
     sherpa-onnx-cxx-api
     portaudio_static
   )
+
+  add_executable(zipformer-transducer-simulate-streaming-microphone-cxx-api
+    ./zipformer-transducer-simulate-streaming-microphone-cxx-api.cc
+    ${CMAKE_CURRENT_LIST_DIR}/../sherpa-onnx/csrc/microphone.cc
+  )
+  target_link_libraries(zipformer-transducer-simulate-streaming-microphone-cxx-api
+    sherpa-onnx-cxx-api
+    portaudio_static
+  )
 endif()
 
 if(SHERPA_ONNX_HAS_ALSA)

cxx-api-examples/nemo-canary-cxx-api.cc

@@ -0,0 +1,101 @@
+// cxx-api-examples/nemo-canary-cxx-api.cc
+//
+// Copyright (c)  2025  Xiaomi Corporation
+
+//
+// This file demonstrates how to use NeMo Canary models with
+// sherpa-onnx's C++ API.
+//
+// clang-format off
+//
+// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+// tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+// rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+//
+// clang-format on
+//
+// see https://k2-fsa.github.io/sherpa/onnx/nemo/canary.html
+// for details
+
+#include <chrono>  // NOLINT
+#include <iostream>
+#include <string>
+
+#include "sherpa-onnx/c-api/cxx-api.h"
+
+int32_t main() {
+  using namespace sherpa_onnx::cxx;  // NOLINT
+  OfflineRecognizerConfig config;
+
+  config.model_config.canary.encoder =
+      "sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx";
+  config.model_config.canary.decoder =
+      "sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx";
+
+  // our input audio is German, so we set src_lang to "de"
+  config.model_config.canary.src_lang = "de";
+
+  // we can set tgt_lang either to de or en in this specific case
+  config.model_config.canary.tgt_lang = "en";
+  config.model_config.tokens =
+      "sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt";
+
+  config.model_config.num_threads = 1;
+
+  std::cout << "Loading model\n";
+  OfflineRecognizer recognizer = OfflineRecognizer::Create(config);
+  if (!recognizer.Get()) {
+    std::cerr << "Please check your config\n";
+    return -1;
+  }
+  std::cout << "Loading model done\n";
+
+  std::string wave_filename =
+      "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/de.wav";
+
+  Wave wave = ReadWave(wave_filename);
+  if (wave.samples.empty()) {
+    std::cerr << "Failed to read: '" << wave_filename << "'\n";
+    return -1;
+  }
+
+  std::cout << "Start recognition\n";
+  const auto begin = std::chrono::steady_clock::now();
+
+  OfflineStream stream = recognizer.CreateStream();
+  stream.AcceptWaveform(wave.sample_rate, wave.samples.data(),
+                        wave.samples.size());
+
+  recognizer.Decode(&stream);
+
+  OfflineRecognizerResult result = recognizer.GetResult(&stream);
+
+  const auto end = std::chrono::steady_clock::now();
+  const float elapsed_seconds =
+      std::chrono::duration_cast<std::chrono::milliseconds>(end - begin)
+          .count() /
+      1000.;
+  float duration = wave.samples.size() / static_cast<float>(wave.sample_rate);
+  float rtf = elapsed_seconds / duration;
+
+  std::cout << "text (English): " << result.text << "\n";
+  printf("Number of threads: %d\n", config.model_config.num_threads);
+  printf("Duration: %.3fs\n", duration);
+  printf("Elapsed seconds: %.3fs\n", elapsed_seconds);
+  printf("(Real time factor) RTF = %.3f / %.3f = %.3f\n", elapsed_seconds,
+         duration, rtf);
+
+  // now output text in German
+  config.model_config.canary.tgt_lang = "de";
+  recognizer.SetConfig(config);
+  stream = recognizer.CreateStream();
+  stream.AcceptWaveform(wave.sample_rate, wave.samples.data(),
+                        wave.samples.size());
+
+  recognizer.Decode(&stream);
+
+  result = recognizer.GetResult(&stream);
+  std::cout << "text (German): " << result.text << "\n";
+
+  return 0;
+}

cxx-api-examples/parakeet-tdt-ctc-simulate-streaming-microphone-cxx-api.cc

@@ -0,0 +1,238 @@
+// cxx-api-examples/parakeet-tdt-simulate-streaming-microphone-cxx-api.cc
+// Copyright (c)  2025  Xiaomi Corporation
+
+//
+// This file demonstrates how to use parakeet-tdt with sherpa-onnx's C++ API
+// for streaming speech recognition from a microphone.
+//
+// clang-format off
+//
+// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_vad.onnx
+//
+// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-parakeet-tdt_ctc-0.6b-ja-35000-int8.tar.bz2
+// tar xvf sherpa-onnx-nemo-parakeet-tdt_ctc-0.6b-ja-35000-int8.tar.bz2
+// rm sherpa-onnx-nemo-parakeet-tdt_ctc-0.6b-ja-35000-int8.tar.bz2
+//
+// clang-format on
+
+#include <signal.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <chrono>              // NOLINT
+#include <condition_variable>  // NOLINT
+#include <iostream>
+#include <mutex>  // NOLINT
+#include <queue>
+#include <vector>
+
+#include "portaudio.h"       // NOLINT
+#include "sherpa-display.h"  // NOLINT
+#include "sherpa-onnx/c-api/cxx-api.h"
+#include "sherpa-onnx/csrc/microphone.h"
+
+std::queue<std::vector<float>> samples_queue;
+std::condition_variable condition_variable;
+std::mutex mutex;
+bool stop = false;
+
+static void Handler(int32_t /*sig*/) {
+  stop = true;
+  condition_variable.notify_one();
+  fprintf(stderr, "\nCaught Ctrl + C. Exiting...\n");
+}
+
+static int32_t RecordCallback(const void *input_buffer,
+                              void * /*output_buffer*/,
+                              unsigned long frames_per_buffer,  // NOLINT
+                              const PaStreamCallbackTimeInfo * /*time_info*/,
+                              PaStreamCallbackFlags /*status_flags*/,
+                              void * /*user_data*/) {
+  std::lock_guard<std::mutex> lock(mutex);
+  samples_queue.emplace(
+      reinterpret_cast<const float *>(input_buffer),
+      reinterpret_cast<const float *>(input_buffer) + frames_per_buffer);
+  condition_variable.notify_one();
+
+  return stop ? paComplete : paContinue;
+}
+
+static sherpa_onnx::cxx::VoiceActivityDetector CreateVad() {
+  using namespace sherpa_onnx::cxx;  // NOLINT
+  VadModelConfig config;
+  config.silero_vad.model = "./silero_vad.onnx";
+  config.silero_vad.threshold = 0.25;
+  config.silero_vad.min_silence_duration = 0.25;
+  config.silero_vad.min_speech_duration = 0.25;
+  config.silero_vad.max_speech_duration = 5;
+  config.sample_rate = 16000;
+  config.debug = false;
+
+  VoiceActivityDetector vad = VoiceActivityDetector::Create(config, 60);
+  if (!vad.Get()) {
+    std::cerr << "Failed to create VAD. Please check your config\n";
+    exit(-1);
+  }
+
+  return vad;
+}
+
+static sherpa_onnx::cxx::OfflineRecognizer CreateOfflineRecognizer() {
+  using namespace sherpa_onnx::cxx;  // NOLINT
+  OfflineRecognizerConfig config;
+
+  config.model_config.nemo_ctc.model =
+      "./sherpa-onnx-nemo-parakeet-tdt_ctc-0.6b-ja-35000-int8/model.int8.onnx";
+  config.model_config.tokens =
+      "./sherpa-onnx-nemo-parakeet-tdt_ctc-0.6b-ja-35000-int8/tokens.txt";
+
+  config.model_config.num_threads = 2;
+  config.model_config.debug = false;
+
+  std::cout << "Loading model\n";
+  OfflineRecognizer recognizer = OfflineRecognizer::Create(config);
+  if (!recognizer.Get()) {
+    std::cerr << "Please check your config\n";
+    exit(-1);
+  }
+  std::cout << "Loading model done\n";
+  return recognizer;
+}
+
+int32_t main() {
+  signal(SIGINT, Handler);
+
+  using namespace sherpa_onnx::cxx;  // NOLINT
+
+  auto vad = CreateVad();
+  auto recognizer = CreateOfflineRecognizer();
+
+  sherpa_onnx::Microphone mic;
+
+  PaDeviceIndex num_devices = Pa_GetDeviceCount();
+  if (num_devices == 0) {
+    std::cerr << "  If you are using Linux, please try to modify "
+                 "./build/bin/sense-voice-simulate-streaming-alsa-cxx-api\n";
+    return -1;
+  }
+
+  int32_t device_index = Pa_GetDefaultInputDevice();
+  const char *pDeviceIndex = std::getenv("SHERPA_ONNX_MIC_DEVICE");
+  if (pDeviceIndex) {
+    fprintf(stderr, "Use specified device: %s\n", pDeviceIndex);
+    device_index = atoi(pDeviceIndex);
+  }
+  mic.PrintDevices(device_index);
+
+  float mic_sample_rate = 16000;
+  const char *sample_rate_str = std::getenv("SHERPA_ONNX_MIC_SAMPLE_RATE");
+  if (sample_rate_str) {
+    mic_sample_rate = atof(sample_rate_str);
+    fprintf(stderr, "Use sample rate %f for mic\n", mic_sample_rate);
+  }
+
+  float sample_rate = 16000;
+  LinearResampler resampler;
+  if (mic_sample_rate != sample_rate) {
+    float min_freq = std::min(mic_sample_rate, sample_rate);
+    float lowpass_cutoff = 0.99 * 0.5 * min_freq;
+
+    int32_t lowpass_filter_width = 6;
+    resampler = LinearResampler::Create(mic_sample_rate, sample_rate,
+                                        lowpass_cutoff, lowpass_filter_width);
+  }
+  if (!mic.OpenDevice(device_index, mic_sample_rate, 1, RecordCallback,
+                      nullptr)) {
+    std::cerr << "Failed to open microphone device\n";
+    return -1;
+  }
+
+  int32_t window_size = 512;  // samples, please don't change
+
+  int32_t offset = 0;
+  std::vector<float> buffer;
+  bool speech_started = false;
+
+  auto started_time = std::chrono::steady_clock::now();
+
+  SherpaDisplay display;
+
+  std::cout << "Started! Please speak\n";
+
+  while (!stop) {
+    {
+      std::unique_lock<std::mutex> lock(mutex);
+      while (samples_queue.empty() && !stop) {
+        condition_variable.wait(lock);
+      }
+
+      const auto &s = samples_queue.front();
+      if (!resampler.Get()) {
+        buffer.insert(buffer.end(), s.begin(), s.end());
+      } else {
+        auto resampled = resampler.Resample(s.data(), s.size(), false);
+        buffer.insert(buffer.end(), resampled.begin(), resampled.end());
+      }
+
+      samples_queue.pop();
+    }
+
+    for (; offset + window_size < buffer.size(); offset += window_size) {
+      vad.AcceptWaveform(buffer.data() + offset, window_size);
+      if (!speech_started && vad.IsDetected()) {
+        speech_started = true;
+        started_time = std::chrono::steady_clock::now();
+      }
+    }
+    if (!speech_started) {
+      if (buffer.size() > 10 * window_size) {
+        offset -= buffer.size() - 10 * window_size;
+        buffer = {buffer.end() - 10 * window_size, buffer.end()};
+      }
+    }
+
+    auto current_time = std::chrono::steady_clock::now();
+    const float elapsed_seconds =
+        std::chrono::duration_cast<std::chrono::milliseconds>(current_time -
+                                                              started_time)
+            .count() /
+        1000.;
+
+    if (speech_started && elapsed_seconds > 0.2) {
+      OfflineStream stream = recognizer.CreateStream();
+      stream.AcceptWaveform(sample_rate, buffer.data(), buffer.size());
+
+      recognizer.Decode(&stream);
+
+      OfflineRecognizerResult result = recognizer.GetResult(&stream);
+      display.UpdateText(result.text);
+      display.Display();
+
+      started_time = std::chrono::steady_clock::now();
+    }
+
+    while (!vad.IsEmpty()) {
+      auto segment = vad.Front();
+
+      vad.Pop();
+
+      OfflineStream stream = recognizer.CreateStream();
+      stream.AcceptWaveform(sample_rate, segment.samples.data(),
+                            segment.samples.size());
+
+      recognizer.Decode(&stream);
+
+      OfflineRecognizerResult result = recognizer.GetResult(&stream);
+
+      display.UpdateText(result.text);
+      display.FinalizeCurrentSentence();
+      display.Display();
+
+      buffer.clear();
+      offset = 0;
+      speech_started = false;
+    }
+  }
+
+  return 0;
+}

cxx-api-examples/parakeet-tdt-simulate-streaming-microphone-cxx-api.cc

@@ -136,11 +136,7 @@ int32_t main() {
     fprintf(stderr, "Use sample rate %f for mic\n", mic_sample_rate);
     mic_sample_rate = atof(sample_rate_str);
   }
-  if (!mic.OpenDevice(device_index, mic_sample_rate, 1, RecordCallback,
+
-                      nullptr) == false) {
-    std::cerr << "Failed to open microphone device\n";
-    return -1;
-  }
   float sample_rate = 16000;
   LinearResampler resampler;
   if (mic_sample_rate != sample_rate) {
@@ -152,6 +148,12 @@ int32_t main() {
                                         lowpass_cutoff, lowpass_filter_width);
   }
 
+  if (!mic.OpenDevice(device_index, mic_sample_rate, 1, RecordCallback,
+                      nullptr)) {
+    std::cerr << "Failed to open microphone device\n";
+    return -1;
+  }
+
   int32_t window_size = 512;  // samples, please don't change
 
   int32_t offset = 0;

cxx-api-examples/sense-voice-simulate-streaming-microphone-cxx-api.cc

@@ -142,8 +142,8 @@ int32_t main() {
     resampler = LinearResampler::Create(mic_sample_rate, sample_rate,
                                         lowpass_cutoff, lowpass_filter_width);
   }
-  if (mic.OpenDevice(device_index, mic_sample_rate, 1, RecordCallback,
+  if (!mic.OpenDevice(device_index, mic_sample_rate, 1, RecordCallback,
-                     nullptr) == false) {
+                      nullptr)) {
     std::cerr << "Failed to open microphone device\n";
     return -1;
   }

cxx-api-examples/zipformer-ctc-simulate-streaming-microphone-cxx-api.cc

@@ -140,8 +140,8 @@ int32_t main() {
     resampler = LinearResampler::Create(mic_sample_rate, sample_rate,
                                         lowpass_cutoff, lowpass_filter_width);
   }
-  if (mic.OpenDevice(device_index, mic_sample_rate, 1, RecordCallback,
+  if (!mic.OpenDevice(device_index, mic_sample_rate, 1, RecordCallback,
-                     nullptr) == false) {
+                      nullptr)) {
     std::cerr << "Failed to open microphone device\n";
     return -1;
   }

cxx-api-examples/zipformer-transducer-simulate-streaming-microphone-cxx-api.cc

@@ -0,0 +1,245 @@
+// cxx-api-examples/zipformer-transducer-simulate-streaming-microphone-cxx-api.cc
+// Copyright (c)  2025  Xiaomi Corporation
+//
+// This file demonstrates how to use Zipformer transducer with sherpa-onnx's C++ API
+// for streaming speech recognition from a microphone.
+//
+// clang-format off
+//
+// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_vad.onnx
+//
+// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01.tar.bz2
+// tar xvf sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01.tar.bz2
+// rm sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01.tar.bz2
+//
+// clang-format on
+
+#include <signal.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <chrono>              // NOLINT
+#include <condition_variable>  // NOLINT
+#include <iostream>
+#include <mutex>  // NOLINT
+#include <queue>
+#include <vector>
+
+#include "portaudio.h"       // NOLINT
+#include "sherpa-display.h"  // NOLINT
+#include "sherpa-onnx/c-api/cxx-api.h"
+#include "sherpa-onnx/csrc/microphone.h"
+
+std::queue<std::vector<float>> samples_queue;
+std::condition_variable condition_variable;
+std::mutex mutex;
+bool stop = false;
+
+static void Handler(int32_t /*sig*/) {
+  stop = true;
+  condition_variable.notify_one();
+  fprintf(stderr, "\nCaught Ctrl + C. Exiting...\n");
+}
+
+static int32_t RecordCallback(const void *input_buffer,
+                              void * /*output_buffer*/,
+                              unsigned long frames_per_buffer,  // NOLINT
+                              const PaStreamCallbackTimeInfo * /*time_info*/,
+                              PaStreamCallbackFlags /*status_flags*/,
+                              void * /*user_data*/) {
+  std::lock_guard<std::mutex> lock(mutex);
+  samples_queue.emplace(
+      reinterpret_cast<const float *>(input_buffer),
+      reinterpret_cast<const float *>(input_buffer) + frames_per_buffer);
+  condition_variable.notify_one();
+
+  return stop ? paComplete : paContinue;
+}
+
+static sherpa_onnx::cxx::VoiceActivityDetector CreateVad() {
+  using namespace sherpa_onnx::cxx;  // NOLINT
+  VadModelConfig config;
+  config.silero_vad.model = "./silero_vad.onnx";
+  config.silero_vad.threshold = 0.5;
+  config.silero_vad.min_silence_duration = 0.1;
+  config.silero_vad.min_speech_duration = 0.25;
+  config.silero_vad.max_speech_duration = 8;
+  config.sample_rate = 16000;
+  config.debug = false;
+
+  VoiceActivityDetector vad = VoiceActivityDetector::Create(config, 20);
+  if (!vad.Get()) {
+    std::cerr << "Failed to create VAD. Please check your config\n";
+    exit(-1);
+  }
+
+  return vad;
+}
+
+static sherpa_onnx::cxx::OfflineRecognizer CreateOfflineRecognizer() {
+  using namespace sherpa_onnx::cxx;  // NOLINT
+  OfflineRecognizerConfig config;
+
+  config.model_config.transducer.encoder =
+      "./sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01/"
+      "encoder-epoch-99-avg-1.int8.onnx";
+
+  config.model_config.transducer.decoder =
+      "./sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01/"
+      "decoder-epoch-99-avg-1.onnx";
+
+  config.model_config.transducer.joiner =
+      "./sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01/"
+      "joiner-epoch-99-avg-1.int8.onnx";
+  config.model_config.tokens =
+      "./sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01/tokens.txt";
+
+  config.model_config.num_threads = 2;
+  config.model_config.debug = false;
+
+  std::cout << "Loading model\n";
+  OfflineRecognizer recognizer = OfflineRecognizer::Create(config);
+  if (!recognizer.Get()) {
+    std::cerr << "Please check your config\n";
+    exit(-1);
+  }
+  std::cout << "Loading model done\n";
+  return recognizer;
+}
+
+int32_t main() {
+  signal(SIGINT, Handler);
+
+  using namespace sherpa_onnx::cxx;  // NOLINT
+
+  auto vad = CreateVad();
+  auto recognizer = CreateOfflineRecognizer();
+
+  sherpa_onnx::Microphone mic;
+
+  PaDeviceIndex num_devices = Pa_GetDeviceCount();
+  if (num_devices == 0) {
+    std::cerr << "  If you are using Linux, please try "
+                 "./build/bin/zipformer-ctc-simulate-streaming-alsa-cxx-api\n";
+    return -1;
+  }
+
+  int32_t device_index = Pa_GetDefaultInputDevice();
+  const char *pDeviceIndex = std::getenv("SHERPA_ONNX_MIC_DEVICE");
+  if (pDeviceIndex) {
+    fprintf(stderr, "Use specified device: %s\n", pDeviceIndex);
+    device_index = atoi(pDeviceIndex);
+  }
+  mic.PrintDevices(device_index);
+
+  float mic_sample_rate = 16000;
+  const char *sample_rate_str = std::getenv("SHERPA_ONNX_MIC_SAMPLE_RATE");
+  if (sample_rate_str) {
+    fprintf(stderr, "Use sample rate %f for mic\n", mic_sample_rate);
+    mic_sample_rate = atof(sample_rate_str);
+  }
+  float sample_rate = 16000;
+  LinearResampler resampler;
+  if (mic_sample_rate != sample_rate) {
+    float min_freq = std::min(mic_sample_rate, sample_rate);
+    float lowpass_cutoff = 0.99 * 0.5 * min_freq;
+
+    int32_t lowpass_filter_width = 6;
+    resampler = LinearResampler::Create(mic_sample_rate, sample_rate,
+                                        lowpass_cutoff, lowpass_filter_width);
+  }
+  if (!mic.OpenDevice(device_index, mic_sample_rate, 1, RecordCallback,
+                      nullptr)) {
+    std::cerr << "Failed to open microphone device\n";
+    return -1;
+  }
+
+  int32_t window_size = 512;  // samples, please don't change
+
+  int32_t offset = 0;
+  std::vector<float> buffer;
+  bool speech_started = false;
+
+  auto started_time = std::chrono::steady_clock::now();
+
+  SherpaDisplay display;
+
+  std::cout << "Started! Please speak\n";
+
+  while (!stop) {
+    {
+      std::unique_lock<std::mutex> lock(mutex);
+      while (samples_queue.empty() && !stop) {
+        condition_variable.wait(lock);
+      }
+
+      const auto &s = samples_queue.front();
+      if (!resampler.Get()) {
+        buffer.insert(buffer.end(), s.begin(), s.end());
+      } else {
+        auto resampled = resampler.Resample(s.data(), s.size(), false);
+        buffer.insert(buffer.end(), resampled.begin(), resampled.end());
+      }
+
+      samples_queue.pop();
+    }
+
+    for (; offset + window_size < buffer.size(); offset += window_size) {
+      vad.AcceptWaveform(buffer.data() + offset, window_size);
+      if (!speech_started && vad.IsDetected()) {
+        speech_started = true;
+        started_time = std::chrono::steady_clock::now();
+      }
+    }
+    if (!speech_started) {
+      if (buffer.size() > 10 * window_size) {
+        offset -= buffer.size() - 10 * window_size;
+        buffer = {buffer.end() - 10 * window_size, buffer.end()};
+      }
+    }
+
+    auto current_time = std::chrono::steady_clock::now();
+    const float elapsed_seconds =
+        std::chrono::duration_cast<std::chrono::milliseconds>(current_time -
+                                                              started_time)
+            .count() /
+        1000.;
+
+    if (speech_started && elapsed_seconds > 0.2) {
+      OfflineStream stream = recognizer.CreateStream();
+      stream.AcceptWaveform(sample_rate, buffer.data(), buffer.size());
+
+      recognizer.Decode(&stream);
+
+      OfflineRecognizerResult result = recognizer.GetResult(&stream);
+      display.UpdateText(result.text);
+      display.Display();
+
+      started_time = std::chrono::steady_clock::now();
+    }
+
+    while (!vad.IsEmpty()) {
+      auto segment = vad.Front();
+
+      vad.Pop();
+
+      OfflineStream stream = recognizer.CreateStream();
+      stream.AcceptWaveform(sample_rate, segment.samples.data(),
+                            segment.samples.size());
+
+      recognizer.Decode(&stream);
+
+      OfflineRecognizerResult result = recognizer.GetResult(&stream);
+
+      display.UpdateText(result.text);
+      display.FinalizeCurrentSentence();
+      display.Display();
+
+      buffer.clear();
+      offset = 0;
+      speech_started = false;
+    }
+  }
+
+  return 0;
+}

dart-api-examples/add-punctuations/pubspec.yaml

@@ -9,7 +9,7 @@ environment:
   sdk: ">=3.0.0 <4.0.0"
 
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   path: ^1.9.0
   args: ^2.5.0
 

dart-api-examples/audio-tagging/pubspec.yaml

@@ -9,7 +9,7 @@ environment:
   sdk: ">=3.0.0 <4.0.0"
 
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   path: ^1.9.0
   args: ^2.5.0
 

dart-api-examples/keyword-spotter/pubspec.yaml

@@ -9,7 +9,7 @@ environment:
   sdk: ">=3.0.0 <4.0.0"
 
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   # sherpa_onnx:
   #   path: ../../flutter/sherpa_onnx
   path: ^1.9.0

dart-api-examples/non-streaming-asr/bin/nemo-canary.dart

@@ -0,0 +1,84 @@
+// Copyright (c)  2025  Xiaomi Corporation
+import 'dart:io';
+
+import 'package:args/args.dart';
+import 'package:sherpa_onnx/sherpa_onnx.dart' as sherpa_onnx;
+
+import './init.dart';
+
+void main(List<String> arguments) async {
+  await initSherpaOnnx();
+
+  final parser = ArgParser()
+    ..addOption('encoder', help: 'Path to the NeMo Canary encoder model')
+    ..addOption('decoder', help: 'Path to the NeMo Canary decoder model')
+    ..addOption('src-lang', help: 'Language of the input audio')
+    ..addOption('tgt-lang', help: 'Language of the recognition result')
+    ..addOption('tokens', help: 'Path to tokens.txt')
+    ..addOption('input-wav', help: 'Path to input.wav to transcribe');
+
+  final res = parser.parse(arguments);
+  if (res['encoder'] == null ||
+      res['decoder'] == null ||
+      res['src-lang'] == null ||
+      res['tgt-lang'] == null ||
+      res['tokens'] == null ||
+      res['input-wav'] == null) {
+    print(parser.usage);
+    exit(1);
+  }
+
+  final encoder = res['encoder'] as String;
+  final decoder = res['decoder'] as String;
+  final srcLang = res['src-lang'] as String;
+  final tgtLang = res['tgt-lang'] as String;
+  final tokens = res['tokens'] as String;
+  final inputWav = res['input-wav'] as String;
+
+  final canary = sherpa_onnx.OfflineCanaryModelConfig(
+      encoder: encoder, decoder: decoder, srcLang: srcLang, tgtLang: tgtLang);
+
+  final modelConfig = sherpa_onnx.OfflineModelConfig(
+    canary: canary,
+    tokens: tokens,
+    debug: false,
+    numThreads: 1,
+  );
+  var config = sherpa_onnx.OfflineRecognizerConfig(model: modelConfig);
+  final recognizer = sherpa_onnx.OfflineRecognizer(config);
+
+  final waveData = sherpa_onnx.readWave(inputWav);
+  final stream = recognizer.createStream();
+
+  stream.acceptWaveform(
+      samples: waveData.samples, sampleRate: waveData.sampleRate);
+  recognizer.decode(stream);
+
+  final result = recognizer.getResult(stream);
+  print('Result in $tgtLang: ${result.text}');
+
+  stream.free();
+
+  // Example to change the target language to de
+  if (tgtLang != 'en') {
+    var json = config.toJson();
+
+    ((json['model'] as Map<String, dynamic>)!['canary']
+        as Map<String, dynamic>)!['tgtLang'] = 'en';
+
+    config = sherpa_onnx.OfflineRecognizerConfig.fromJson(json);
+    recognizer.setConfig(config);
+
+    final stream = recognizer.createStream();
+
+    stream.acceptWaveform(
+        samples: waveData.samples, sampleRate: waveData.sampleRate);
+    recognizer.decode(stream);
+
+    final result = recognizer.getResult(stream);
+    print('Result in English: ${result.text}');
+    stream.free();
+  }
+
+  recognizer.free();
+}

dart-api-examples/non-streaming-asr/pubspec.yaml

@@ -10,7 +10,7 @@ environment:
 
 # Add regular dependencies here.
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   path: ^1.9.0
   args: ^2.5.0
 

dart-api-examples/non-streaming-asr/run-nemo-canary.sh

@@ -0,0 +1,33 @@
+#!/usr/bin/env bash
+
+set -ex
+
+dart pub get
+
+if [ ! -f sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+fi
+
+for tgt_lang in en de es fr; do
+  dart run \
+    ./bin/nemo-canary.dart \
+    --encoder ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx \
+    --decoder ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx \
+    --tokens ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt \
+    --src-lang en \
+    --tgt-lang $tgt_lang \
+    --input-wav ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/en.wav
+done
+
+for tgt_lang in en de; do
+  dart run \
+    ./bin/nemo-canary.dart \
+    --encoder ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx \
+    --decoder ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx \
+    --tokens ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt \
+    --src-lang de \
+    --tgt-lang $tgt_lang \
+    --input-wav ./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/de.wav
+done

dart-api-examples/speaker-diarization/pubspec.yaml

@@ -8,7 +8,7 @@ environment:
   sdk: ">=3.0.0 <4.0.0"
 
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   # sherpa_onnx:
   #   path: ../../flutter/sherpa_onnx
   path: ^1.9.0

dart-api-examples/speaker-identification/pubspec.yaml

@@ -9,7 +9,7 @@ environment:
   sdk: ">=3.0.0 <4.0.0"
 
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   path: ^1.9.0
   args: ^2.5.0
 

dart-api-examples/speech-enhancement-gtcrn/pubspec.yaml

@@ -10,7 +10,7 @@ environment:
 
 # Add regular dependencies here.
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   # sherpa_onnx:
   #   path: ../../flutter/sherpa_onnx
   path: ^1.9.0

dart-api-examples/streaming-asr/pubspec.yaml

@@ -11,7 +11,7 @@ environment:
 
 # Add regular dependencies here.
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   path: ^1.9.0
   args: ^2.5.0
 

dart-api-examples/tts/pubspec.yaml

@@ -8,7 +8,7 @@ environment:
 
 # Add regular dependencies here.
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   path: ^1.9.0
   args: ^2.5.0
 

dart-api-examples/vad-with-non-streaming-asr/pubspec.yaml

@@ -10,7 +10,7 @@ environment:
   sdk: ">=3.0.0 <4.0.0"
 
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   path: ^1.9.0
   args: ^2.5.0
 

dart-api-examples/vad/pubspec.yaml

@@ -9,7 +9,7 @@ environment:
   sdk: ">=3.0.0 <4.0.0"
 
 dependencies:
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   path: ^1.9.0
   args: ^2.5.0
 

dotnet-examples/non-streaming-canary-decode-files/Program.cs

@@ -0,0 +1,44 @@
+// Copyright (c)  2025  Xiaomi Corporation
+//
+// This file shows how to use a NeMo Canary model for speech recognition.
+//
+// You can find the model doc at
+// https://k2-fsa.github.io/sherpa/onnx/nemo/canary.html
+using SherpaOnnx;
+
+class NonStreamingAsrCanary
+{
+  static void Main(string[] args)
+  {
+    // please download model files from
+    // https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
+    var config = new OfflineRecognizerConfig();
+    config.ModelConfig.Canary.Encoder = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx";
+    config.ModelConfig.Canary.Decoder = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx";
+    config.ModelConfig.Canary.SrcLang = "en";
+    config.ModelConfig.Canary.TgtLang = "en";
+    config.ModelConfig.Tokens = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt";
+    config.ModelConfig.Debug = 0;
+    var recognizer = new OfflineRecognizer(config);
+
+    var testWaveFilename = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/en.wav";
+    var reader = new WaveReader(testWaveFilename);
+    var stream = recognizer.CreateStream();
+    stream.AcceptWaveform(reader.SampleRate, reader.Samples);
+    recognizer.Decode(stream);
+    var text = stream.Result.Text;
+    Console.WriteLine("Text (English): {0}", text);
+
+    // Now output text in German
+    config.ModelConfig.Canary.TgtLang = "de";
+    recognizer.SetConfig(config);
+
+    stream = recognizer.CreateStream();
+    stream.AcceptWaveform(reader.SampleRate, reader.Samples);
+    recognizer.Decode(stream);
+    text = stream.Result.Text;
+    Console.WriteLine("Text (German): {0}", text);
+  }
+}
+
+

dotnet-examples/non-streaming-canary-decode-files/non-streaming-canary-decode-files.csproj

@@ -0,0 +1,15 @@
+<Project Sdk="Microsoft.NET.Sdk">
+
+  <PropertyGroup>
+    <OutputType>Exe</OutputType>
+    <TargetFramework>net8.0</TargetFramework>
+    <RootNamespace>non_streaming_canary_decode_files</RootNamespace>
+    <ImplicitUsings>enable</ImplicitUsings>
+    <Nullable>enable</Nullable>
+  </PropertyGroup>
+
+  <ItemGroup>
+    <ProjectReference Include="..\Common\Common.csproj" />
+  </ItemGroup>
+
+</Project>

dotnet-examples/non-streaming-canary-decode-files/run.sh

@@ -0,0 +1,11 @@
+#!/usr/bin/env bash
+
+set -ex
+
+if [ ! -f sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+fi
+
+dotnet run

dotnet-examples/sherpa-onnx.sln

@@ -39,6 +39,8 @@ Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "speech-enhancement-gtcrn",
 EndProject
 Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "version-test", "version-test\version-test.csproj", "{E57711E5-6546-4BA0-B627-79C94F415BC5}"
 EndProject
+Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "non-streaming-canary-decode-files", "non-streaming-canary-decode-files\non-streaming-canary-decode-files.csproj", "{925779DB-4429-4366-87C3-B14DD44AE1D4}"
+EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
 		Debug|Any CPU = Debug|Any CPU
@@ -117,6 +119,10 @@ Global
 		{E57711E5-6546-4BA0-B627-79C94F415BC5}.Debug|Any CPU.Build.0 = Debug|Any CPU
 		{E57711E5-6546-4BA0-B627-79C94F415BC5}.Release|Any CPU.ActiveCfg = Release|Any CPU
 		{E57711E5-6546-4BA0-B627-79C94F415BC5}.Release|Any CPU.Build.0 = Release|Any CPU
+		{925779DB-4429-4366-87C3-B14DD44AE1D4}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
+		{925779DB-4429-4366-87C3-B14DD44AE1D4}.Debug|Any CPU.Build.0 = Debug|Any CPU
+		{925779DB-4429-4366-87C3-B14DD44AE1D4}.Release|Any CPU.ActiveCfg = Release|Any CPU
+		{925779DB-4429-4366-87C3-B14DD44AE1D4}.Release|Any CPU.Build.0 = Release|Any CPU
 	EndGlobalSection
 	GlobalSection(SolutionProperties) = preSolution
 		HideSolutionNode = FALSE

flutter-examples/streaming_asr/pubspec.yaml

@@ -5,7 +5,7 @@ description: >
 
 publish_to: 'none'
 
-version: 1.12.4
+version: 1.12.5
 
 topics:
   - speech-recognition
@@ -31,7 +31,7 @@ dependencies:
   record: ^5.1.0
   url_launcher: ^6.2.6
 
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   # sherpa_onnx:
   #   path: ../../flutter/sherpa_onnx
 

flutter-examples/tts/pubspec.yaml

@@ -5,7 +5,7 @@ description: >
 
 publish_to: 'none' # Remove this line if you wish to publish to pub.dev
 
-version: 1.12.4
+version: 1.12.5
 
 environment:
   sdk: ">=2.17.0 <4.0.0"
@@ -18,7 +18,7 @@ dependencies:
   cupertino_icons: ^1.0.6
   path_provider: ^2.1.3
   path: ^1.9.0
-  sherpa_onnx: ^1.12.4
+  sherpa_onnx: ^1.12.5
   # sherpa_onnx:
   #   path: ../../flutter/sherpa_onnx
   url_launcher: 6.2.6

flutter/sherpa_onnx/lib/src/offline_recognizer.dart

@@ -163,6 +163,44 @@ class OfflineWhisperModelConfig {
   final int tailPaddings;
 }
 
+class OfflineCanaryModelConfig {
+  const OfflineCanaryModelConfig(
+      {this.encoder = '',
+      this.decoder = '',
+      this.srcLang = 'en',
+      this.tgtLang = 'en',
+      this.usePnc = true});
+
+  factory OfflineCanaryModelConfig.fromJson(Map<String, dynamic> json) {
+    return OfflineCanaryModelConfig(
+      encoder: json['encoder'] as String? ?? '',
+      decoder: json['decoder'] as String? ?? '',
+      srcLang: json['srcLang'] as String? ?? 'en',
+      tgtLang: json['tgtLang'] as String? ?? 'en',
+      usePnc: json['usePnc'] as bool? ?? true,
+    );
+  }
+
+  @override
+  String toString() {
+    return 'OfflineCanaryModelConfig(encoder: $encoder, decoder: $decoder, srcLang: $srcLang, tgtLang: $tgtLang, usePnc: $usePnc)';
+  }
+
+  Map<String, dynamic> toJson() => {
+        'encoder': encoder,
+        'decoder': decoder,
+        'srcLang': srcLang,
+        'tgtLang': tgtLang,
+        'usePnc': usePnc,
+      };
+
+  final String encoder;
+  final String decoder;
+  final String srcLang;
+  final String tgtLang;
+  final bool usePnc;
+}
+
 class OfflineFireRedAsrModelConfig {
   const OfflineFireRedAsrModelConfig({this.encoder = '', this.decoder = ''});
 
@@ -310,6 +348,7 @@ class OfflineModelConfig {
     this.fireRedAsr = const OfflineFireRedAsrModelConfig(),
     this.dolphin = const OfflineDolphinModelConfig(),
     this.zipformerCtc = const OfflineZipformerCtcModelConfig(),
+    this.canary = const OfflineCanaryModelConfig(),
     required this.tokens,
     this.numThreads = 1,
     this.debug = true,
@@ -362,6 +401,10 @@ class OfflineModelConfig {
           ? OfflineZipformerCtcModelConfig.fromJson(
               json['zipformerCtc'] as Map<String, dynamic>)
           : const OfflineZipformerCtcModelConfig(),
+      canary: json['canary'] != null
+          ? OfflineCanaryModelConfig.fromJson(
+              json['canary'] as Map<String, dynamic>)
+          : const OfflineCanaryModelConfig(),
       tokens: json['tokens'] as String,
       numThreads: json['numThreads'] as int? ?? 1,
       debug: json['debug'] as bool? ?? true,
@@ -375,7 +418,7 @@ class OfflineModelConfig {
 
   @override
   String toString() {
-    return 'OfflineModelConfig(transducer: $transducer, paraformer: $paraformer, nemoCtc: $nemoCtc, whisper: $whisper, tdnn: $tdnn, senseVoice: $senseVoice, moonshine: $moonshine, fireRedAsr: $fireRedAsr, dolphin: $dolphin, zipformerCtc: $zipformerCtc, tokens: $tokens, numThreads: $numThreads, debug: $debug, provider: $provider, modelType: $modelType, modelingUnit: $modelingUnit, bpeVocab: $bpeVocab, telespeechCtc: $telespeechCtc)';
+    return 'OfflineModelConfig(transducer: $transducer, paraformer: $paraformer, nemoCtc: $nemoCtc, whisper: $whisper, tdnn: $tdnn, senseVoice: $senseVoice, moonshine: $moonshine, fireRedAsr: $fireRedAsr, dolphin: $dolphin, zipformerCtc: $zipformerCtc, canary: $canary, tokens: $tokens, numThreads: $numThreads, debug: $debug, provider: $provider, modelType: $modelType, modelingUnit: $modelingUnit, bpeVocab: $bpeVocab, telespeechCtc: $telespeechCtc)';
   }
 
   Map<String, dynamic> toJson() => {
@@ -389,6 +432,7 @@ class OfflineModelConfig {
         'fireRedAsr': fireRedAsr.toJson(),
         'dolphin': dolphin.toJson(),
         'zipformerCtc': zipformerCtc.toJson(),
+        'canary': canary.toJson(),
         'tokens': tokens,
         'numThreads': numThreads,
         'debug': debug,
@@ -409,6 +453,7 @@ class OfflineModelConfig {
   final OfflineFireRedAsrModelConfig fireRedAsr;
   final OfflineDolphinModelConfig dolphin;
   final OfflineZipformerCtcModelConfig zipformerCtc;
+  final OfflineCanaryModelConfig canary;
 
   final String tokens;
   final int numThreads;
@@ -549,7 +594,28 @@ class OfflineRecognizer {
 
   /// The user is responsible to call the OfflineRecognizer.free()
   /// method of the returned instance to avoid memory leak.
+
   factory OfflineRecognizer(OfflineRecognizerConfig config) {
+    final c = convertConfig(config);
+
+    final ptr = SherpaOnnxBindings.createOfflineRecognizer?.call(c) ?? nullptr;
+
+    freeConfig(c);
+
+    return OfflineRecognizer._(ptr: ptr, config: config);
+  }
+
+  void setConfig(OfflineRecognizerConfig config) {
+    final c = convertConfig(config);
+
+    SherpaOnnxBindings.offlineRecognizerSetConfig?.call(ptr, c);
+
+    freeConfig(c);
+    // we don't update this.config
+  }
+
+  static Pointer<SherpaOnnxOfflineRecognizerConfig> convertConfig(
+      OfflineRecognizerConfig config) {
     final c = calloc<SherpaOnnxOfflineRecognizerConfig>();
 
     c.ref.feat.sampleRate = config.feat.sampleRate;
@@ -609,6 +675,12 @@ class OfflineRecognizer {
     c.ref.model.zipformerCtc.model =
         config.model.zipformerCtc.model.toNativeUtf8();
 
+    c.ref.model.canary.encoder = config.model.canary.encoder.toNativeUtf8();
+    c.ref.model.canary.decoder = config.model.canary.decoder.toNativeUtf8();
+    c.ref.model.canary.srcLang = config.model.canary.srcLang.toNativeUtf8();
+    c.ref.model.canary.tgtLang = config.model.canary.tgtLang.toNativeUtf8();
+    c.ref.model.canary.usePnc = config.model.canary.usePnc ? 1 : 0;
+
     c.ref.model.tokens = config.model.tokens.toNativeUtf8();
 
     c.ref.model.numThreads = config.model.numThreads;
@@ -637,8 +709,10 @@ class OfflineRecognizer {
     c.ref.hr.lexicon = config.hr.lexicon.toNativeUtf8();
     c.ref.hr.ruleFsts = config.hr.ruleFsts.toNativeUtf8();
 
-    final ptr = SherpaOnnxBindings.createOfflineRecognizer?.call(c) ?? nullptr;
+    return c;
+  }
 
+  static void freeConfig(Pointer<SherpaOnnxOfflineRecognizerConfig> c) {
     calloc.free(c.ref.hr.dictDir);
     calloc.free(c.ref.hr.lexicon);
     calloc.free(c.ref.hr.ruleFsts);
@@ -653,6 +727,10 @@ class OfflineRecognizer {
     calloc.free(c.ref.model.modelType);
     calloc.free(c.ref.model.provider);
     calloc.free(c.ref.model.tokens);
+    calloc.free(c.ref.model.canary.tgtLang);
+    calloc.free(c.ref.model.canary.srcLang);
+    calloc.free(c.ref.model.canary.decoder);
+    calloc.free(c.ref.model.canary.encoder);
     calloc.free(c.ref.model.zipformerCtc.model);
     calloc.free(c.ref.model.dolphin.model);
     calloc.free(c.ref.model.fireRedAsr.decoder);
@@ -674,8 +752,6 @@ class OfflineRecognizer {
     calloc.free(c.ref.model.transducer.decoder);
     calloc.free(c.ref.model.transducer.joiner);
     calloc.free(c);
-
-    return OfflineRecognizer._(ptr: ptr, config: config);
   }
 
   /// The user has to invoke stream.free() on the returned instance

flutter/sherpa_onnx/lib/src/sherpa_onnx_bindings.dart

@@ -280,6 +280,16 @@ final class SherpaOnnxOfflineWhisperModelConfig extends Struct {
   external int tailPaddings;
 }
 
+final class SherpaOnnxOfflineCanaryModelConfig extends Struct {
+  external Pointer<Utf8> encoder;
+  external Pointer<Utf8> decoder;
+  external Pointer<Utf8> srcLang;
+  external Pointer<Utf8> tgtLang;
+
+  @Int32()
+  external int usePnc;
+}
+
 final class SherpaOnnxOfflineMoonshineModelConfig extends Struct {
   external Pointer<Utf8> preprocessor;
   external Pointer<Utf8> encoder;
@@ -338,6 +348,7 @@ final class SherpaOnnxOfflineModelConfig extends Struct {
   external SherpaOnnxOfflineFireRedAsrModelConfig fireRedAsr;
   external SherpaOnnxOfflineDolphinModelConfig dolphin;
   external SherpaOnnxOfflineZipformerCtcModelConfig zipformerCtc;
+  external SherpaOnnxOfflineCanaryModelConfig canary;
 }
 
 final class SherpaOnnxOfflineRecognizerConfig extends Struct {
@@ -876,6 +887,14 @@ typedef CreateOfflineRecognizerNative = Pointer<SherpaOnnxOfflineRecognizer>
 
 typedef CreateOfflineRecognizer = CreateOfflineRecognizerNative;
 
+typedef OfflineRecognizerSetConfigNative = Void Function(
+    Pointer<SherpaOnnxOfflineRecognizer>,
+    Pointer<SherpaOnnxOfflineRecognizerConfig>);
+
+typedef OfflineRecognizerSetConfig = void Function(
+    Pointer<SherpaOnnxOfflineRecognizer>,
+    Pointer<SherpaOnnxOfflineRecognizerConfig>);
+
 typedef DestroyOfflineRecognizerNative = Void Function(
     Pointer<SherpaOnnxOfflineRecognizer>);
 
@@ -1341,6 +1360,7 @@ class SherpaOnnxBindings {
 
   static CreateOfflineRecognizer? createOfflineRecognizer;
   static DestroyOfflineRecognizer? destroyOfflineRecognizer;
+  static OfflineRecognizerSetConfig? offlineRecognizerSetConfig;
   static CreateOfflineStream? createOfflineStream;
   static DestroyOfflineStream? destroyOfflineStream;
   static AcceptWaveformOffline? acceptWaveformOffline;
@@ -1741,6 +1761,11 @@ class SherpaOnnxBindings {
             'SherpaOnnxDestroyOfflineRecognizer')
         .asFunction();
 
+    offlineRecognizerSetConfig ??= dynamicLibrary
+        .lookup<NativeFunction<OfflineRecognizerSetConfigNative>>(
+            'SherpaOnnxOfflineRecognizerSetConfig')
+        .asFunction();
+
     createOfflineStream ??= dynamicLibrary
         .lookup<NativeFunction<CreateOfflineStreamNative>>(
             'SherpaOnnxCreateOfflineStream')

flutter/sherpa_onnx/pubspec.yaml

@@ -17,7 +17,7 @@ topics:
   - voice-activity-detection
 
 # remember to change the version in ../sherpa_onnx_macos/macos/sherpa_onnx_macos.podspec
-version: 1.12.4
+version: 1.12.5
 
 homepage: https://github.com/k2-fsa/sherpa-onnx
 
@@ -30,23 +30,23 @@ dependencies:
   flutter:
     sdk: flutter
 
-  sherpa_onnx_android: ^1.12.4
+  sherpa_onnx_android: ^1.12.5
   # sherpa_onnx_android:
   #   path: ../sherpa_onnx_android
 
-  sherpa_onnx_macos: ^1.12.4
+  sherpa_onnx_macos: ^1.12.5
   # sherpa_onnx_macos:
   #   path: ../sherpa_onnx_macos
 
-  sherpa_onnx_linux: ^1.12.4
+  sherpa_onnx_linux: ^1.12.5
   # sherpa_onnx_linux:
   #   path: ../sherpa_onnx_linux
 
-  sherpa_onnx_windows: ^1.12.4
+  sherpa_onnx_windows: ^1.12.5
   # sherpa_onnx_windows:
   #   path: ../sherpa_onnx_windows
 
-  sherpa_onnx_ios: ^1.12.4
+  sherpa_onnx_ios: ^1.12.5
   # sherpa_onnx_ios:
   #   path: ../sherpa_onnx_ios
 

flutter/sherpa_onnx_ios/ios/sherpa_onnx_ios.podspec

@@ -7,7 +7,7 @@
 # https://groups.google.com/g/dart-ffi/c/nUATMBy7r0c
 Pod::Spec.new do |s|
   s.name             = 'sherpa_onnx_ios'
-  s.version          = '1.12.4'
+  s.version          = '1.12.5'
   s.summary          = 'A new Flutter FFI plugin project.'
   s.description      = <<-DESC
 A new Flutter FFI plugin project.

flutter/sherpa_onnx_macos/macos/sherpa_onnx_macos.podspec

@@ -4,7 +4,7 @@
 #
 Pod::Spec.new do |s|
   s.name             = 'sherpa_onnx_macos'
-  s.version          = '1.12.4'
+  s.version          = '1.12.5'
   s.summary          = 'sherpa-onnx Flutter FFI plugin project.'
   s.description      = <<-DESC
 sherpa-onnx Flutter FFI plugin project.

go-api-examples/non-streaming-canary-decode-files/go.mod

@@ -0,0 +1,17 @@
+module non-streaming-canary-decode-files
+
+go 1.17
+
+require (
+	github.com/k2-fsa/sherpa-onnx-go v1.12.4
+	github.com/spf13/pflag v1.0.6
+	github.com/youpy/go-wav v0.3.2
+)
+
+require (
+	github.com/k2-fsa/sherpa-onnx-go-linux v1.12.4 // indirect
+	github.com/k2-fsa/sherpa-onnx-go-macos v1.12.4 // indirect
+	github.com/k2-fsa/sherpa-onnx-go-windows v1.12.4 // indirect
+	github.com/youpy/go-riff v0.1.0 // indirect
+	github.com/zaf/g711 v0.0.0-20190814101024-76a4a538f52b // indirect
+)

go-api-examples/non-streaming-canary-decode-files/main.go

@@ -0,0 +1,113 @@
+package main
+
+import (
+	"bytes"
+	"encoding/binary"
+	"log"
+	"os"
+	"strings"
+
+	sherpa "github.com/k2-fsa/sherpa-onnx-go/sherpa_onnx"
+	"github.com/youpy/go-wav"
+)
+
+func main() {
+	log.SetFlags(log.LstdFlags | log.Lmicroseconds)
+
+	config := sherpa.OfflineRecognizerConfig{}
+
+	config.ModelConfig.Canary.Encoder = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx"
+	config.ModelConfig.Canary.Decoder = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx"
+	config.ModelConfig.Canary.SrcLang = "en"
+	config.ModelConfig.Canary.TgtLang = "en"
+	config.ModelConfig.Canary.UsePnc = 1
+	config.ModelConfig.Tokens = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt"
+
+	waveFilename := "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/en.wav"
+
+	samples, sampleRate := readWave(waveFilename)
+
+	log.Println("Initializing recognizer (may take several seconds)")
+	recognizer := sherpa.NewOfflineRecognizer(&config)
+	log.Println("Recognizer created!")
+	defer sherpa.DeleteOfflineRecognizer(recognizer)
+
+	log.Println("Start decoding!")
+	stream := sherpa.NewOfflineStream(recognizer)
+	defer sherpa.DeleteOfflineStream(stream)
+
+	stream.AcceptWaveform(sampleRate, samples)
+
+	recognizer.Decode(stream)
+	log.Println("Decoding done!")
+	result := stream.GetResult()
+
+	log.Println("Text in English: " + strings.ToLower(result.Text))
+
+	s := sherpa.NewOfflineStream(recognizer)
+	defer sherpa.DeleteOfflineStream(s)
+
+	s.AcceptWaveform(sampleRate, samples)
+
+	config.ModelConfig.Canary.TgtLang = "de"
+	recognizer.SetConfig(&config)
+	recognizer.Decode(s)
+	result = s.GetResult()
+
+	log.Println("Text in German: " + strings.ToLower(result.Text))
+}
+
+func readWave(filename string) (samples []float32, sampleRate int) {
+	file, _ := os.Open(filename)
+	defer file.Close()
+
+	reader := wav.NewReader(file)
+	format, err := reader.Format()
+	if err != nil {
+		log.Fatalf("Failed to read wave format")
+	}
+
+	if format.AudioFormat != 1 {
+		log.Fatalf("Support only PCM format. Given: %v\n", format.AudioFormat)
+	}
+
+	if format.NumChannels != 1 {
+		log.Fatalf("Support only 1 channel wave file. Given: %v\n", format.NumChannels)
+	}
+
+	if format.BitsPerSample != 16 {
+		log.Fatalf("Support only 16-bit per sample. Given: %v\n", format.BitsPerSample)
+	}
+
+	reader.Duration() // so that it initializes reader.Size
+
+	buf := make([]byte, reader.Size)
+	n, err := reader.Read(buf)
+	if n != int(reader.Size) {
+		log.Fatalf("Failed to read %v bytes. Returned %v bytes\n", reader.Size, n)
+	}
+
+	samples = samplesInt16ToFloat(buf)
+	sampleRate = int(format.SampleRate)
+
+	return
+}
+
+func samplesInt16ToFloat(inSamples []byte) []float32 {
+	numSamples := len(inSamples) / 2
+	outSamples := make([]float32, numSamples)
+
+	for i := 0; i != numSamples; i++ {
+		s := inSamples[i*2 : (i+1)*2]
+
+		var s16 int16
+		buf := bytes.NewReader(s)
+		err := binary.Read(buf, binary.LittleEndian, &s16)
+		if err != nil {
+			log.Fatal("Failed to parse 16-bit sample")
+		}
+		outSamples[i] = float32(s16) / 32768
+	}
+
+	return outSamples
+}

go-api-examples/non-streaming-canary-decode-files/run.sh

@@ -0,0 +1,13 @@
+#!/usr/bin/env bash
+
+set -ex
+
+if [ ! -f sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+fi
+
+go mod tidy
+go build
+./non-streaming-canary-decode-files

harmony-os/SherpaOnnxHar/sherpa_onnx/BuildProfile.ets

@@ -1,7 +1,7 @@
 /**
  * Use these variables when you tailor your ArkTS code. They must be of the const type.
  */
-export const HAR_VERSION = '1.12.4';
+export const HAR_VERSION = '1.12.5';
 export const BUILD_MODE_NAME = 'debug';
 export const DEBUG = true;
 export const TARGET_NAME = 'default';

harmony-os/SherpaOnnxHar/sherpa_onnx/Index.ets

@@ -7,6 +7,7 @@ export { Samples,
   OfflineStream,
   FeatureConfig,
   HomophoneReplacerConfig,
+  OfflineCanaryModelConfig,
   OfflineDolphinModelConfig,
   OfflineTransducerModelConfig,
   OfflineParaformerModelConfig,

harmony-os/SherpaOnnxHar/sherpa_onnx/README.md

@@ -23,7 +23,7 @@ or update your `oh-package.json5` to include the following:
 
 ```
   "dependencies": {
-    "sherpa_onnx": "1.12.4",
+    "sherpa_onnx": "1.12.5",
   },
 ```
 

harmony-os/SherpaOnnxHar/sherpa_onnx/oh-package.json5

@@ -1,6 +1,6 @@
 {
   "name": "sherpa_onnx",
-  "version": "1.12.4",
+  "version": "1.12.5",
   "description": "On-device speech-to-text, text-to-speech, and speaker diarization using Next-gen Kaldi without Internet connection",
   "main": "Index.ets",
   "author": "The next-gen Kaldi team",

harmony-os/SherpaOnnxHar/sherpa_onnx/src/main/cpp/non-streaming-asr.cc

@@ -93,6 +93,27 @@ static SherpaOnnxOfflineNemoEncDecCtcModelConfig GetOfflineNeMoCtcModelConfig(
   return c;
 }
 
+static SherpaOnnxOfflineCanaryModelConfig GetOfflineCanaryModelConfig(
+    Napi::Object obj) {
+  SherpaOnnxOfflineCanaryModelConfig c;
+  memset(&c, 0, sizeof(c));
+  c.use_pnc = 1;  // Align default with JS default
+
+  if (!obj.Has("canary") || !obj.Get("canary").IsObject()) {
+    return c;
+  }
+
+  Napi::Object o = obj.Get("canary").As<Napi::Object>();
+
+  SHERPA_ONNX_ASSIGN_ATTR_STR(encoder, encoder);
+  SHERPA_ONNX_ASSIGN_ATTR_STR(decoder, decoder);
+  SHERPA_ONNX_ASSIGN_ATTR_STR(src_lang, srcLang);
+  SHERPA_ONNX_ASSIGN_ATTR_STR(tgt_lang, tgtLang);
+  SHERPA_ONNX_ASSIGN_ATTR_INT32(use_pnc, usePnc);
+
+  return c;
+}
+
 static SherpaOnnxOfflineWhisperModelConfig GetOfflineWhisperModelConfig(
     Napi::Object obj) {
   SherpaOnnxOfflineWhisperModelConfig c;
@@ -203,6 +224,7 @@ static SherpaOnnxOfflineModelConfig GetOfflineModelConfig(Napi::Object obj) {
   c.fire_red_asr = GetOfflineFireRedAsrModelConfig(o);
   c.dolphin = GetOfflineDolphinModelConfig(o);
   c.zipformer_ctc = GetOfflineZipformerCtcModelConfig(o);
+  c.canary = GetOfflineCanaryModelConfig(o);
 
   SHERPA_ONNX_ASSIGN_ATTR_STR(tokens, tokens);
   SHERPA_ONNX_ASSIGN_ATTR_INT32(num_threads, numThreads);
@@ -241,6 +263,78 @@ static SherpaOnnxOfflineLMConfig GetOfflineLMConfig(Napi::Object obj) {
   return c;
 }
 
+static SherpaOnnxOfflineRecognizerConfig ParseConfig(Napi::Object o) {
+  SherpaOnnxOfflineRecognizerConfig c;
+  memset(&c, 0, sizeof(c));
+  c.feat_config = GetFeatureConfig(o);
+  c.model_config = GetOfflineModelConfig(o);
+  c.lm_config = GetOfflineLMConfig(o);
+  c.hr = GetHomophoneReplacerConfig(o);
+
+  SHERPA_ONNX_ASSIGN_ATTR_STR(decoding_method, decodingMethod);
+  SHERPA_ONNX_ASSIGN_ATTR_INT32(max_active_paths, maxActivePaths);
+  SHERPA_ONNX_ASSIGN_ATTR_STR(hotwords_file, hotwordsFile);
+  SHERPA_ONNX_ASSIGN_ATTR_FLOAT(hotwords_score, hotwordsScore);
+  SHERPA_ONNX_ASSIGN_ATTR_STR(rule_fsts, ruleFsts);
+  SHERPA_ONNX_ASSIGN_ATTR_STR(rule_fars, ruleFars);
+  SHERPA_ONNX_ASSIGN_ATTR_FLOAT(blank_penalty, blankPenalty);
+
+  return c;
+}
+
+static void FreeConfig(const SherpaOnnxOfflineRecognizerConfig &c) {
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.transducer.encoder);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.transducer.decoder);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.transducer.joiner);
+
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.paraformer.model);
+
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.nemo_ctc.model);
+
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.whisper.encoder);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.whisper.decoder);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.whisper.language);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.whisper.task);
+
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.tdnn.model);
+
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.sense_voice.model);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.sense_voice.language);
+
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.moonshine.preprocessor);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.moonshine.encoder);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.moonshine.uncached_decoder);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.moonshine.cached_decoder);
+
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.fire_red_asr.encoder);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.fire_red_asr.decoder);
+
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.dolphin.model);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.zipformer_ctc.model);
+
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.canary.encoder);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.canary.decoder);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.canary.src_lang);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.canary.tgt_lang);
+
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.tokens);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.provider);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.model_type);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.modeling_unit);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.bpe_vocab);
+  SHERPA_ONNX_DELETE_C_STR(c.model_config.telespeech_ctc);
+
+  SHERPA_ONNX_DELETE_C_STR(c.lm_config.model);
+
+  SHERPA_ONNX_DELETE_C_STR(c.decoding_method);
+  SHERPA_ONNX_DELETE_C_STR(c.hotwords_file);
+  SHERPA_ONNX_DELETE_C_STR(c.rule_fsts);
+  SHERPA_ONNX_DELETE_C_STR(c.rule_fars);
+  SHERPA_ONNX_DELETE_C_STR(c.hr.dict_dir);
+  SHERPA_ONNX_DELETE_C_STR(c.hr.lexicon);
+  SHERPA_ONNX_DELETE_C_STR(c.hr.rule_fsts);
+}
+
 static Napi::External<SherpaOnnxOfflineRecognizer>
 CreateOfflineRecognizerWrapper(const Napi::CallbackInfo &info) {
   Napi::Env env = info.Env();
@@ -274,20 +368,7 @@ CreateOfflineRecognizerWrapper(const Napi::CallbackInfo &info) {
 
   Napi::Object o = info[0].As<Napi::Object>();
 
-  SherpaOnnxOfflineRecognizerConfig c;
+  SherpaOnnxOfflineRecognizerConfig c = ParseConfig(o);
-  memset(&c, 0, sizeof(c));
-  c.feat_config = GetFeatureConfig(o);
-  c.model_config = GetOfflineModelConfig(o);
-  c.lm_config = GetOfflineLMConfig(o);
-  c.hr = GetHomophoneReplacerConfig(o);
-
-  SHERPA_ONNX_ASSIGN_ATTR_STR(decoding_method, decodingMethod);
-  SHERPA_ONNX_ASSIGN_ATTR_INT32(max_active_paths, maxActivePaths);
-  SHERPA_ONNX_ASSIGN_ATTR_STR(hotwords_file, hotwordsFile);
-  SHERPA_ONNX_ASSIGN_ATTR_FLOAT(hotwords_score, hotwordsScore);
-  SHERPA_ONNX_ASSIGN_ATTR_STR(rule_fsts, ruleFsts);
-  SHERPA_ONNX_ASSIGN_ATTR_STR(rule_fars, ruleFars);
-  SHERPA_ONNX_ASSIGN_ATTR_FLOAT(blank_penalty, blankPenalty);
 
 #if __OHOS__
   std::unique_ptr<NativeResourceManager,
@@ -302,51 +383,7 @@ CreateOfflineRecognizerWrapper(const Napi::CallbackInfo &info) {
       SherpaOnnxCreateOfflineRecognizer(&c);
 #endif
 
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.transducer.encoder);
+  FreeConfig(c);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.transducer.decoder);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.transducer.joiner);
-
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.paraformer.model);
-
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.nemo_ctc.model);
-
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.whisper.encoder);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.whisper.decoder);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.whisper.language);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.whisper.task);
-
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.tdnn.model);
-
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.sense_voice.model);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.sense_voice.language);
-
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.moonshine.preprocessor);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.moonshine.encoder);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.moonshine.uncached_decoder);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.moonshine.cached_decoder);
-
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.fire_red_asr.encoder);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.fire_red_asr.decoder);
-
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.dolphin.model);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.zipformer_ctc.model);
-
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.tokens);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.provider);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.model_type);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.modeling_unit);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.bpe_vocab);
-  SHERPA_ONNX_DELETE_C_STR(c.model_config.telespeech_ctc);
-
-  SHERPA_ONNX_DELETE_C_STR(c.lm_config.model);
-
-  SHERPA_ONNX_DELETE_C_STR(c.decoding_method);
-  SHERPA_ONNX_DELETE_C_STR(c.hotwords_file);
-  SHERPA_ONNX_DELETE_C_STR(c.rule_fsts);
-  SHERPA_ONNX_DELETE_C_STR(c.rule_fars);
-  SHERPA_ONNX_DELETE_C_STR(c.hr.dict_dir);
-  SHERPA_ONNX_DELETE_C_STR(c.hr.lexicon);
-  SHERPA_ONNX_DELETE_C_STR(c.hr.rule_fsts);
 
   if (!recognizer) {
     Napi::TypeError::New(env, "Please check your config!")
@@ -470,6 +507,43 @@ static void AcceptWaveformOfflineWrapper(const Napi::CallbackInfo &info) {
 #endif
 }
 
+static void OfflineRecognizerSetConfigWrapper(const Napi::CallbackInfo &info) {
+  Napi::Env env = info.Env();
+  if (info.Length() != 2) {
+    std::ostringstream os;
+    os << "Expect only 2 arguments. Given: " << info.Length();
+
+    Napi::TypeError::New(env, os.str()).ThrowAsJavaScriptException();
+
+    return;
+  }
+
+  if (!info[0].IsExternal()) {
+    Napi::TypeError::New(env,
+                         "Argument 0 should be an offline recognizer pointer.")
+        .ThrowAsJavaScriptException();
+
+    return;
+  }
+
+  if (!info[1].IsObject()) {
+    Napi::TypeError::New(env, "Expect an object as the second argument")
+        .ThrowAsJavaScriptException();
+
+    return;
+  }
+
+  Napi::Object o = info[1].As<Napi::Object>();
+  SherpaOnnxOfflineRecognizerConfig c = ParseConfig(o);
+
+  const SherpaOnnxOfflineRecognizer *recognizer =
+      info[0].As<Napi::External<SherpaOnnxOfflineRecognizer>>().Data();
+
+  SherpaOnnxOfflineRecognizerSetConfig(recognizer, &c);
+
+  FreeConfig(c);
+}
+
 static void DecodeOfflineStreamWrapper(const Napi::CallbackInfo &info) {
   Napi::Env env = info.Env();
   if (info.Length() != 2) {
@@ -548,6 +622,9 @@ void InitNonStreamingAsr(Napi::Env env, Napi::Object exports) {
   exports.Set(Napi::String::New(env, "decodeOfflineStream"),
               Napi::Function::New(env, DecodeOfflineStreamWrapper));
 
+  exports.Set(Napi::String::New(env, "offlineRecognizerSetConfig"),
+              Napi::Function::New(env, OfflineRecognizerSetConfigWrapper));
+
   exports.Set(Napi::String::New(env, "getOfflineStreamResultAsJson"),
               Napi::Function::New(env, GetOfflineStreamResultAsJsonWrapper));
 }

harmony-os/SherpaOnnxHar/sherpa_onnx/src/main/cpp/types/libsherpa_onnx/Index.d.ts

@@ -22,6 +22,7 @@ export const voiceActivityDetectorFlush: (handle: object) => void;
 
 export const createOfflineRecognizer: (config: object, mgr?: object) => object;
 export const createOfflineStream: (handle: object) => object;
+export const offlineRecognizerSetConfig: (handle: object, config: object) => void;
 export const acceptWaveformOffline: (handle: object, audio: object) => void;
 export const decodeOfflineStream: (handle: object, streamHandle: object) => void;
 export const getOfflineStreamResultAsJson: (streamHandle: object) => string;

harmony-os/SherpaOnnxHar/sherpa_onnx/src/main/ets/components/NonStreamingAsr.ets

@@ -4,6 +4,7 @@ import {
   createOfflineStream,
   decodeOfflineStream,
   getOfflineStreamResultAsJson,
+  offlineRecognizerSetConfig,
 } from 'libsherpa_onnx.so';
 
 export interface Samples {
@@ -67,6 +68,14 @@ export class OfflineWhisperModelConfig {
   public tailPaddings: number = -1;
 }
 
+export class OfflineCanaryModelConfig {
+  public encoder: string = '';
+  public decoder: string = '';
+  public srcLang: string = '';
+  public tgtLang: string = '';
+  public usePnc: number = 1;
+}
+
 export class OfflineTdnnModelConfig {
   public model: string = '';
 }
@@ -102,6 +111,7 @@ export class OfflineModelConfig {
   public moonshine: OfflineMoonshineModelConfig = new OfflineMoonshineModelConfig();
   public dolphin: OfflineDolphinModelConfig = new OfflineDolphinModelConfig();
   public zipformerCtc: OfflineZipformerCtcModelConfig = new OfflineZipformerCtcModelConfig();
+  public canary: OfflineCanaryModelConfig = new OfflineCanaryModelConfig();
 }
 
 export class OfflineLMConfig {
@@ -151,6 +161,10 @@ export class OfflineRecognizer {
     this.config = config
   }
 
+  setConfig(config: OfflineRecognizerConfig) {
+    offlineRecognizerSetConfig(this.handle, config);
+  }
+
   createStream(): OfflineStream {
     const handle: object = createOfflineStream(this.handle);
     return new OfflineStream(handle);

harmony-os/SherpaOnnxSpeakerDiarization/entry/oh-package.json5

@@ -6,7 +6,7 @@
   "author": "",
   "license": "",
   "dependencies": {
-    "sherpa_onnx": "1.12.4"
+    "sherpa_onnx": "1.12.5"
   }
 }
 

harmony-os/SherpaOnnxSpeakerIdentification/entry/oh-package.json5

@@ -6,7 +6,7 @@
   "author": "",
   "license": "",
   "dependencies": {
-    "sherpa_onnx": "1.12.4",
+    "sherpa_onnx": "1.12.5",
   }
 }
 

harmony-os/SherpaOnnxStreamingAsr/entry/oh-package.json5

@@ -6,7 +6,7 @@
   "author": "",
   "license": "",
   "dependencies": {
-    "sherpa_onnx": "1.12.4",
+    "sherpa_onnx": "1.12.5",
   }
 }
 

harmony-os/SherpaOnnxTts/entry/oh-package.json5

@@ -6,7 +6,7 @@
   "author": "",
   "license": "",
   "dependencies": {
-    "sherpa_onnx": "1.12.4",
+    "sherpa_onnx": "1.12.5",
   }
 }
 

harmony-os/SherpaOnnxVadAsr/entry/README.md

@@ -1,6 +1,6 @@
 # Introduction
 
-Please download ./sherpa_onnx-v1.12.4.har
+Please download ./sherpa_onnx-v1.12.5.har
 from <https://huggingface.co/csukuangfj/sherpa-onnx-harmony-os/tree/main/har>
 
 Hint: For users who have no access to huggingface, please use

harmony-os/SherpaOnnxVadAsr/entry/oh-package.json5

@@ -7,7 +7,7 @@
   "license": "",
   "dependencies": {
     // please see https://ohpm.openharmony.cn/#/cn/detail/sherpa_onnx
-    "sherpa_onnx": "1.12.4",
+    "sherpa_onnx": "1.12.5",
   }
 }
 

java-api-examples/NonStreamingDecodeFileNemoCanary.java

@@ -0,0 +1,56 @@
+// Copyright 2024 Xiaomi Corporation
+
+// This file shows how to use an offline NeMo Canary model, i.e.,
+// non-streaming NeMo Canary model, to decode files.
+import com.k2fsa.sherpa.onnx.*;
+
+public class NonStreamingDecodeFileNemoCanary {
+  public static void main(String[] args) {
+    // please refer to
+    // https://k2-fsa.github.io/sherpa/onnx/nemo/canary.html
+    // to download model files
+    String encoder = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx";
+    String decoder = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx";
+    String tokens = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt";
+
+    String waveFilename = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/en.wav";
+
+    WaveReader reader = new WaveReader(waveFilename);
+
+    OfflineCanaryModelConfig canary =
+        OfflineCanaryModelConfig.builder()
+            .setEncoder(encoder)
+            .setDecoder(decoder)
+            .setSrcLang("en")
+            .setTgtLang("en")
+            .setUsePnc(true)
+            .build();
+
+    OfflineModelConfig modelConfig =
+        OfflineModelConfig.builder()
+            .setCanary(canary)
+            .setTokens(tokens)
+            .setNumThreads(1)
+            .setDebug(true)
+            .build();
+
+    OfflineRecognizerConfig config =
+        OfflineRecognizerConfig.builder()
+            .setOfflineModelConfig(modelConfig)
+            .setDecodingMethod("greedy_search")
+            .build();
+
+    OfflineRecognizer recognizer = new OfflineRecognizer(config);
+    OfflineStream stream = recognizer.createStream();
+    stream.acceptWaveform(reader.getSamples(), reader.getSampleRate());
+
+    recognizer.decode(stream);
+
+    String text = recognizer.getResult(stream).getText();
+
+    System.out.printf("filename:%s\nresult(English):%s\n", waveFilename, text);
+
+    stream.release();
+    recognizer.release();
+  }
+}

java-api-examples/README.md

@@ -24,11 +24,18 @@ This directory contains examples for the JAVA API of sherpa-onnx.
 
 ```bash
 ./run-non-streaming-decode-file-dolphin-ctc.sh
+./run-non-streaming-decode-file-fire-red-asr.sh
+./run-non-streaming-decode-file-moonshine.sh
+./run-non-streaming-decode-file-nemo-canary.sh
+./run-non-streaming-decode-file-nemo.sh
 ./run-non-streaming-decode-file-paraformer.sh
 ./run-non-streaming-decode-file-sense-voice.sh
+./run-non-streaming-decode-file-tele-speech-ctc.sh
+./run-non-streaming-decode-file-transducer-hotwords.sh
 ./run-non-streaming-decode-file-transducer.sh
+./run-non-streaming-decode-file-whisper-multiple.sh
 ./run-non-streaming-decode-file-whisper.sh
-./run-non-streaming-decode-file-nemo.sh
+./run-non-streaming-decode-file-zipformer-ctc.sh
 ```
 
 ## Non-Streaming Speech recognition with homophone replacer

java-api-examples/run-non-streaming-decode-file-nemo-canary.sh

@@ -0,0 +1,37 @@
+#!/usr/bin/env bash
+
+set -ex
+
+if [[ ! -f ../build/lib/libsherpa-onnx-jni.dylib  && ! -f ../build/lib/libsherpa-onnx-jni.so ]]; then
+  mkdir -p ../build
+  pushd ../build
+  cmake \
+    -DSHERPA_ONNX_ENABLE_PYTHON=OFF \
+    -DSHERPA_ONNX_ENABLE_TESTS=OFF \
+    -DSHERPA_ONNX_ENABLE_CHECK=OFF \
+    -DBUILD_SHARED_LIBS=ON \
+    -DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
+    -DSHERPA_ONNX_ENABLE_JNI=ON \
+    ..
+
+  make -j4
+  ls -lh lib
+  popd
+fi
+
+if [ ! -f ../sherpa-onnx/java-api/build/sherpa-onnx.jar ]; then
+  pushd ../sherpa-onnx/java-api
+  make
+  popd
+fi
+
+if [ ! -f sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+fi
+
+java \
+  -Djava.library.path=$PWD/../build/lib \
+  -cp ../sherpa-onnx/java-api/build/sherpa-onnx.jar \
+  NonStreamingDecodeFileNemoCanary.java

jitpack.yml

@@ -2,8 +2,8 @@ jdk:
   - openjdk17
 
 before_install:
-  - wget https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.4/sherpa-onnx-1.12.4.aar
+  - wget https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.5/sherpa-onnx-1.12.5.aar
 
 install:
-  - FILE="-Dfile=sherpa-onnx-1.12.4.aar"
+  - FILE="-Dfile=sherpa-onnx-1.12.5.aar"
-  - mvn install:install-file $FILE -DgroupId=com.k2fsa.sherpa.onnx -DartifactId=sherpa-onnx -Dversion=1.12.4 -Dpackaging=aar -DgeneratePom=true
+  - mvn install:install-file $FILE -DgroupId=com.k2fsa.sherpa.onnx -DartifactId=sherpa-onnx -Dversion=1.12.5 -Dpackaging=aar -DgeneratePom=true

kotlin-api-examples/run.sh

@@ -455,8 +455,31 @@ function testOfflineSenseVoiceWithHr() {
   ls -lh $out_filename
   java -Djava.library.path=../build/lib -jar $out_filename
 }
-testVersion
 
+function testOfflineNeMoCanary() {
+  if [ ! -f sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx ]; then
+    curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+    tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+    rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  fi
+
+  out_filename=test_offline_nemo_canary.jar
+  kotlinc-jvm -include-runtime -d $out_filename \
+    test_offline_nemo_canary.kt \
+    FeatureConfig.kt \
+    HomophoneReplacerConfig.kt \
+    OfflineRecognizer.kt \
+    OfflineStream.kt \
+    WaveReader.kt \
+    faked-asset-manager.kt
+
+  ls -lh $out_filename
+  java -Djava.library.path=../build/lib -jar $out_filename
+}
+
+# testVersion
+
+testOfflineNeMoCanary
 testOfflineSenseVoiceWithHr
 testOfflineSpeechDenoiser
 testOfflineSpeakerDiarization

kotlin-api-examples/test_offline_nemo_canary.kt

@@ -0,0 +1,48 @@
+package com.k2fsa.sherpa.onnx
+
+fun main() {
+  val recognizer = createOfflineRecognizer()
+  val waveFilename = "./sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/en.wav"
+
+  val objArray = WaveReader.readWaveFromFile(
+      filename = waveFilename,
+  )
+  val samples: FloatArray = objArray[0] as FloatArray
+  val sampleRate: Int = objArray[1] as Int
+
+  var stream = recognizer.createStream()
+  stream.acceptWaveform(samples, sampleRate=sampleRate)
+  recognizer.decode(stream)
+
+  var result = recognizer.getResult(stream)
+  println("English: $result")
+
+  stream.release()
+
+  // now output text in German
+  val config = recognizer.config.copy(modelConfig=recognizer.config.modelConfig.copy(
+    canary=recognizer.config.modelConfig.canary.copy(
+      tgtLang="de"
+    )
+  ))
+  recognizer.setConfig(config)
+
+  stream = recognizer.createStream()
+  stream.acceptWaveform(samples, sampleRate=sampleRate)
+  recognizer.decode(stream)
+
+  result = recognizer.getResult(stream)
+  println("German: $result")
+
+  stream.release()
+  recognizer.release()
+}
+
+
+fun createOfflineRecognizer(): OfflineRecognizer {
+  val config = OfflineRecognizerConfig(
+      modelConfig = getOfflineModelConfig(type = 32)!!,
+  )
+
+  return OfflineRecognizer(config = config)
+}

mfc-examples/README.md

@@ -5,9 +5,9 @@ for speech recognition.
 
 |Directory| Pre-built exe (x64)|Pre-built exe (x86)| Description|
 |---------|--------------------|-------------------|------------|
-|[./NonStreamingSpeechRecognition](./NonStreamingSpeechRecognition)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.4/sherpa-onnx-non-streaming-asr-x64-v1.12.4.exe)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.4/sherpa-onnx-non-streaming-asr-x86-v1.12.4.exe)| Non-streaming speech recognition|
+|[./NonStreamingSpeechRecognition](./NonStreamingSpeechRecognition)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.5/sherpa-onnx-non-streaming-asr-x64-v1.12.5.exe)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.5/sherpa-onnx-non-streaming-asr-x86-v1.12.5.exe)| Non-streaming speech recognition|
-|[./StreamingSpeechRecognition](./StreamingSpeechRecognition)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.4/sherpa-onnx-streaming-asr-x64-v1.12.4.exe)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.4/sherpa-onnx-streaming-asr-x86-v1.12.4.exe)| Streaming speech recognition|
+|[./StreamingSpeechRecognition](./StreamingSpeechRecognition)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.5/sherpa-onnx-streaming-asr-x64-v1.12.5.exe)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.5/sherpa-onnx-streaming-asr-x86-v1.12.5.exe)| Streaming speech recognition|
-|[./NonStreamingTextToSpeech](./NonStreamingTextToSpeech)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.4/sherpa-onnx-non-streaming-tts-x64-v1.12.4.exe)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.4/sherpa-onnx-non-streaming-tts-x86-v1.12.4.exe)| Non-streaming text to speech|
+|[./NonStreamingTextToSpeech](./NonStreamingTextToSpeech)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.5/sherpa-onnx-non-streaming-tts-x64-v1.12.5.exe)|[URL](https://github.com/k2-fsa/sherpa-onnx/releases/download/v1.12.5/sherpa-onnx-non-streaming-tts-x86-v1.12.5.exe)| Non-streaming text to speech|
 
 Caution: You need to use Windows and install Visual Studio 2022 in order to
 compile it.

new-release.sh

@@ -1,8 +1,8 @@
 #!/usr/bin/env bash
 
 set -ex
-old_version="1\.12\.3"
+old_version="1\.12\.4"
-new_version="1\.12\.4"
+new_version="1\.12\.5"
 replace_str="s/$old_version/$new_version/g"
 
 sed -i.bak "$replace_str" ./sherpa-onnx/csrc/version.cc

nodejs-addon-examples/README.md

@@ -123,6 +123,7 @@ The following tables list the examples in this folder.
 |[./test_asr_non_streaming_moonshine.js](./test_asr_non_streaming_moonshine.js)|Non-streaming speech recognition from a file using [Moonshine](https://github.com/usefulsensors/moonshine)|
 |[./test_vad_with_non_streaming_asr_moonshine.js](./test_vad_with_non_streaming_asr_moonshine.js)| Non-streaming speech recognition from a file using [Moonshine](https://github.com/usefulsensors/moonshine) + [Silero VAD](https://github.com/snakers4/silero-vad)|
 |[./test_asr_non_streaming_nemo_ctc.js](./test_asr_non_streaming_nemo_ctc.js)|Non-streaming speech recognition from a file using a [NeMo](https://github.com/NVIDIA/NeMo) CTC model with greedy search|
+|[./test_asr_non_streaming_nemo_canary.js](./test_asr_non_streaming_nemo_canary.js)|Non-streaming speech recognition from a file using a [NeMo](https://github.com/NVIDIA/NeMo) [Canary](https://k2-fsa.github.io/sherpa/onnx/nemo/canary.html#sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8-english-spanish-german-french) model|
 |[./test_asr_non_streaming_zipformer_ctc.js](./test_asr_non_streaming_zipformer_ctc.js)|Non-streaming speech recognition from a file using a Zipformer CTC model with greedy search|
 |[./test_asr_non_streaming_nemo_parakeet_tdt_v2.js](./test_asr_non_streaming_nemo_parakeet_tdt_v2.js)|Non-streaming speech recognition from a file using a [NeMo](https://github.com/NVIDIA/NeMo) [parakeet-tdt-0.6b-v2](https://k2-fsa.github.io/sherpa/onnx/pretrained_models/offline-transducer/nemo-transducer-models.html#sherpa-onnx-nemo-parakeet-tdt-0-6b-v2-int8-english) model with greedy search|
 |[./test_asr_non_streaming_dolphin_ctc.js](./test_asr_non_streaming_dolphin_ctc.js)|Non-streaming speech recognition from a file using a [Dolphinhttps://github.com/DataoceanAI/Dolphin]) CTC model with greedy search|
@@ -389,6 +390,16 @@ npm install naudiodon2
 node ./test_vad_asr_non_streaming_zipformer_ctc_microphone.js
 ```
 
+### Non-streaming speech recognition with NeMo Canary models
+
+```bash
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+
+node ./test_asr_non_streaming_nemo_canary.js
+```
+
 ### Non-streaming speech recognition with NeMo CTC models
 
 ```bash

nodejs-addon-examples/package.json

@@ -1,5 +1,5 @@
 {
   "dependencies": {
-    "sherpa-onnx-node": "^1.12.4"
+    "sherpa-onnx-node": "^1.12.5"
   }
 }

nodejs-addon-examples/test_asr_non_streaming_nemo_canary.js

@@ -0,0 +1,62 @@
+// Copyright (c)  2024  Xiaomi Corporation
+const sherpa_onnx = require('sherpa-onnx-node');
+
+// Please download test files from
+// https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
+const config = {
+  'featConfig': {
+    'sampleRate': 16000,
+    'featureDim': 80,
+  },
+  'modelConfig': {
+    'canary': {
+      'encoder':
+          './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx',
+      'decoder':
+          './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx',
+      'srcLang': 'en',
+      'tgtLang': 'en',
+      'usePnc': 1,
+    },
+    'tokens':
+        './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt',
+    'numThreads': 2,
+    'provider': 'cpu',
+    'debug': 0,
+  }
+};
+
+const waveFilename =
+    './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/en.wav';
+
+const recognizer = new sherpa_onnx.OfflineRecognizer(config);
+console.log('Started')
+let start = Date.now();
+let stream = recognizer.createStream();
+const wave = sherpa_onnx.readWave(waveFilename);
+stream.acceptWaveform({sampleRate: wave.sampleRate, samples: wave.samples});
+
+recognizer.decode(stream);
+let result = recognizer.getResult(stream)
+let stop = Date.now();
+console.log('Done')
+
+const elapsed_seconds = (stop - start) / 1000;
+const duration = wave.samples.length / wave.sampleRate;
+const real_time_factor = elapsed_seconds / duration;
+console.log('Wave duration', duration.toFixed(3), 'seconds')
+console.log('Elapsed', elapsed_seconds.toFixed(3), 'seconds')
+console.log(
+    `RTF = ${elapsed_seconds.toFixed(3)}/${duration.toFixed(3)} =`,
+    real_time_factor.toFixed(3))
+console.log(waveFilename)
+console.log('result (English)\n', result)
+
+stream = recognizer.createStream();
+stream.acceptWaveform({sampleRate: wave.sampleRate, samples: wave.samples});
+recognizer.config.modelConfig.canary.tgtLang = 'de';
+recognizer.setConfig(recognizer.config);
+
+recognizer.decode(stream);
+result = recognizer.getResult(stream)
+console.log('result (German)\n', result)

nodejs-examples/README.md

@@ -63,7 +63,7 @@ for text-to-speech.
 You can use the following command to run it:
 
 ```bash
-wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/kokoro-en-v0_19.tar.bz2
+curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/kokoro-en-v0_19.tar.bz2
 tar xf kokoro-en-v0_19.tar.bz2
 rm kokoro-en-v0_19.tar.bz2
 
@@ -154,6 +154,22 @@ rm sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02.tar.bz2
 node ./test-offline-dolphin-ctc.js
 ```
 
+## ./test-offline-nemo-canary.js
+
+[./test-offline-nemo-canary.js](./test-offline-nemo-canary.js) demonstrates
+how to decode a file with a NeMo Canary model. In the code we use
+[sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8](https://k2-fsa.github.io/sherpa/onnx/nemo/canary.html#sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8-english-spanish-german-french).
+
+You can use the following command to run it:
+
+```bash
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+
+node ./test-offline-nemo-canary.js
+```
+
 ## ./test-offline-zipformer-ctc.js
 
 [./test-offline-zipformer-ctc.js](./test-offline-zipformer-ctc.js) demonstrates

nodejs-examples/package.json

@@ -2,7 +2,7 @@
   "dependencies": {
     "mic": "^2.1.2",
     "naudiodon2": "^2.4.0",
-    "sherpa-onnx": "^1.12.4",
+    "sherpa-onnx": "^1.12.5",
     "wav": "^1.0.2"
   }
 }

nodejs-examples/test-offline-nemo-canary.js

@@ -0,0 +1,56 @@
+// Copyright (c)  2025  Xiaomi Corporation (authors: Fangjun Kuang)
+//
+const fs = require('fs');
+const {Readable} = require('stream');
+const wav = require('wav');
+
+const sherpa_onnx = require('sherpa-onnx');
+
+function createOfflineRecognizer() {
+  let config = {
+    modelConfig: {
+      canary: {
+        encoder:
+            './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx',
+        decoder:
+            './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx',
+        srcLang: 'en',
+        tgtLang: 'en',
+        usePnc: 1,
+      },
+      debug: 0,
+      tokens:
+          './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt',
+    }
+  };
+
+  return sherpa_onnx.createOfflineRecognizer(config);
+}
+
+const recognizer = createOfflineRecognizer();
+let stream = recognizer.createStream();
+
+const waveFilename =
+    './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/en.wav';
+const wave = sherpa_onnx.readWave(waveFilename);
+stream.acceptWaveform(wave.sampleRate, wave.samples);
+
+recognizer.decode(stream);
+let text = recognizer.getResult(stream).text;
+console.log(`text in English: ${text}`);
+
+stream.free();
+
+// now output German text
+recognizer.config.modelConfig.canary.tgtLang = 'de';
+recognizer.setConfig(recognizer.config);
+
+stream = recognizer.createStream();
+stream.acceptWaveform(wave.sampleRate, wave.samples);
+recognizer.decode(stream);
+text = recognizer.getResult(stream).text;
+
+console.log(`text in German: ${text}`);
+
+stream.free();
+recognizer.free();

pascal-api-examples/non-streaming-asr/.gitignore

@@ -10,3 +10,4 @@ telespeech_ctc
 moonshine
 dolphin_ctc
 zipformer_ctc
+nemo_canary

pascal-api-examples/non-streaming-asr/nemo_canary.pas

@@ -0,0 +1,107 @@
+{ Copyright (c)  2025  Xiaomi Corporation }
+
+{
+This file shows how to use a non-streaming NeMo Canary model
+to decode files.
+
+You can download the model files from
+https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
+}
+
+program nemo_canary;
+
+{$mode objfpc}
+
+uses
+  sherpa_onnx,
+  DateUtils,
+  SysUtils;
+
+var
+  Wave: TSherpaOnnxWave;
+  WaveFilename: AnsiString;
+
+  Config: TSherpaOnnxOfflineRecognizerConfig;
+  Recognizer: TSherpaOnnxOfflineRecognizer;
+  Stream: TSherpaOnnxOfflineStream;
+  RecognitionResult: TSherpaOnnxOfflineRecognizerResult;
+
+  Start: TDateTime;
+  Stop: TDateTime;
+
+  Elapsed: Single;
+  Duration: Single;
+  RealTimeFactor: Single;
+begin
+  Initialize(Config);
+
+  Config.ModelConfig.Canary.Encoder := './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx';
+  Config.ModelConfig.Canary.Decoder := './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/decoder.int8.onnx';
+  Config.ModelConfig.Canary.SrcLang := 'en';
+  Config.ModelConfig.Canary.TgtLang := 'en';
+  Config.ModelConfig.Canary.UsePnc := True;
+  Config.ModelConfig.Tokens := './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/tokens.txt';
+  Config.ModelConfig.Provider := 'cpu';
+  Config.ModelConfig.NumThreads := 1;
+  Config.ModelConfig.Debug := False;
+
+  WaveFilename := './sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/test_wavs/en.wav';
+
+  Wave := SherpaOnnxReadWave(WaveFilename);
+
+  Recognizer := TSherpaOnnxOfflineRecognizer.Create(Config);
+  Stream := Recognizer.CreateStream();
+  Start := Now;
+
+  Stream.AcceptWaveform(Wave.Samples, Wave.SampleRate);
+  Recognizer.Decode(Stream);
+
+  RecognitionResult := Recognizer.GetResult(Stream);
+
+  Stop := Now;
+
+  Elapsed := MilliSecondsBetween(Stop, Start) / 1000;
+  Duration := Length(Wave.Samples) / Wave.SampleRate;
+  RealTimeFactor := Elapsed / Duration;
+
+  WriteLn(RecognitionResult.ToString);
+  WriteLn(Format('NumThreads %d', [Config.ModelConfig.NumThreads]));
+  WriteLn(Format('Elapsed %.3f s', [Elapsed]));
+  WriteLn(Format('Wave duration %.3f s', [Duration]));
+  WriteLn(Format('RTF = %.3f/%.3f = %.3f', [Elapsed, Duration, RealTimeFactor]));
+
+  FreeAndNil(Stream);
+
+  WriteLn('-----------Output German-----');
+
+  Stream := Recognizer.CreateStream();
+  Start := Now;
+
+  Stream.AcceptWaveform(Wave.Samples, Wave.SampleRate);
+
+  Config.ModelConfig.Canary.TgtLang := 'de';
+  Recognizer.SetConfig(Config);
+  Recognizer.Decode(Stream);
+
+  RecognitionResult := Recognizer.GetResult(Stream);
+
+  Stop := Now;
+
+  Elapsed := MilliSecondsBetween(Stop, Start) / 1000;
+  Duration := Length(Wave.Samples) / Wave.SampleRate;
+  RealTimeFactor := Elapsed / Duration;
+
+  WriteLn(RecognitionResult.ToString);
+  WriteLn(Format('NumThreads %d', [Config.ModelConfig.NumThreads]));
+  WriteLn(Format('Elapsed %.3f s', [Elapsed]));
+  WriteLn(Format('Wave duration %.3f s', [Duration]));
+  WriteLn(Format('RTF = %.3f/%.3f = %.3f', [Elapsed, Duration, RealTimeFactor]));
+
+  {Free resources to avoid memory leak.
+
+  Note: You don't need to invoke them for this simple script.
+  However, you have to invoke them in your own large/complex project.
+  }
+  FreeAndNil(Stream);
+  FreeAndNil(Recognizer);
+end.

pascal-api-examples/non-streaming-asr/run-nemo-canary.sh

@@ -0,0 +1,42 @@
+#!/usr/bin/env bash
+
+set -ex
+
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
+SHERPA_ONNX_DIR=$(cd $SCRIPT_DIR/../.. && pwd)
+
+echo "SHERPA_ONNX_DIR: $SHERPA_ONNX_DIR"
+
+if [[ ! -f ../../build/install/lib/libsherpa-onnx-c-api.dylib  && ! -f ../../build/install/lib/libsherpa-onnx-c-api.so && ! -f ../../build/install/lib/sherpa-onnx-c-api.dll ]]; then
+  mkdir -p ../../build
+  pushd ../../build
+  cmake \
+    -DCMAKE_INSTALL_PREFIX=./install \
+    -DSHERPA_ONNX_ENABLE_PYTHON=OFF \
+    -DSHERPA_ONNX_ENABLE_TESTS=OFF \
+    -DSHERPA_ONNX_ENABLE_CHECK=OFF \
+    -DBUILD_SHARED_LIBS=ON \
+    -DSHERPA_ONNX_ENABLE_PORTAUDIO=OFF \
+    ..
+
+  cmake --build . --target install --config Release
+  ls -lh lib
+  popd
+fi
+
+if [ ! -f sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8/encoder.int8.onnx ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  tar xvf sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+  rm sherpa-onnx-nemo-canary-180m-flash-en-es-de-fr-int8.tar.bz2
+fi
+
+fpc \
+  -dSHERPA_ONNX_USE_SHARED_LIBS \
+  -Fu$SHERPA_ONNX_DIR/sherpa-onnx/pascal-api \
+  -Fl$SHERPA_ONNX_DIR/build/install/lib \
+  ./nemo_canary.pas
+
+export LD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$LD_LIBRARY_PATH
+export DYLD_LIBRARY_PATH=$SHERPA_ONNX_DIR/build/install/lib:$DYLD_LIBRARY_PATH
+
+./nemo_canary

pom.xml

@@ -4,7 +4,7 @@
     <modelVersion>4.0.0</modelVersion>
     <groupId>com.k2fsa.sherpa.onnx</groupId>
     <artifactId>sherpa-onnx-android</artifactId>
-    <version>1.12.4</version>
+    <version>1.12.5</version>
     <url>https://github.com/k2-fsa/sherpa-onnx</url>
     <packaging>pom</packaging>
     <description>First Android Library</description>

python-api-examples/offline-decode-files.py

@@ -35,6 +35,25 @@ file(s) with a non-streaming model.
       /path/to/0.wav \
       /path/to/1.wav
 
+    also with RNN LM rescoring and LODR (optional):
+
+    ./python-api-examples/offline-decode-files.py  \
+      --tokens=/path/to/tokens.txt \
+      --encoder=/path/to/encoder.onnx \
+      --decoder=/path/to/decoder.onnx \
+      --joiner=/path/to/joiner.onnx \
+      --num-threads=2 \
+      --decoding-method=modified_beam_search \
+      --debug=false \
+      --sample-rate=16000 \
+      --feature-dim=80 \
+      --lm=/path/to/lm.onnx \
+      --lm-scale=0.1 \
+      --lodr-fst=/path/to/lodr.fst \
+      --lodr-scale=-0.1 \
+      /path/to/0.wav \
+      /path/to/1.wav
+
 (3) For CTC models from NeMo
 
 python3 ./python-api-examples/offline-decode-files.py \
@@ -269,6 +288,39 @@ def get_args():
         default="greedy_search",
         help="Valid values are greedy_search and modified_beam_search",
     )
+
+    parser.add_argument(
+        "--lm",
+        metavar="file",
+        type=str,
+        default="",
+        help="Path to RNN LM model",
+    )
+
+    parser.add_argument(
+        "--lm-scale",
+        metavar="lm_scale",
+        type=float,
+        default=0.1,
+        help="LM model scale for rescoring",
+    )
+
+    parser.add_argument(
+        "--lodr-fst",
+        metavar="file",
+        type=str,
+        default="",
+        help="Path to LODR FST model. Used only when --lm is given.",
+    )
+
+    parser.add_argument(
+        "--lodr-scale",
+        metavar="lodr_scale",
+        type=float,
+        default=-0.1,
+        help="LODR scale for rescoring.Used only when --lodr_fst is given.",
+    )
+
     parser.add_argument(
         "--debug",
         type=bool,
@@ -364,6 +416,10 @@ def main():
             num_threads=args.num_threads,
             sample_rate=args.sample_rate,
             feature_dim=args.feature_dim,
+            lm=args.lm,
+            lm_scale=args.lm_scale,
+            lodr_fst=args.lodr_fst,
+            lodr_scale=args.lodr_scale,
             decoding_method=args.decoding_method,
             hotwords_file=args.hotwords_file,
             hotwords_score=args.hotwords_score,

python-api-examples/online-decode-files.py

@@ -21,6 +21,22 @@ rm sherpa-onnx-streaming-zipformer-en-2023-06-26.tar.bz2
   ./sherpa-onnx-streaming-zipformer-en-2023-06-26/test_wavs/1.wav \
   ./sherpa-onnx-streaming-zipformer-en-2023-06-26/test_wavs/8k.wav
 
+or with RNN LM rescoring and LODR:
+
+./python-api-examples/online-decode-files.py \
+  --tokens=./sherpa-onnx-streaming-zipformer-en-2023-06-26/tokens.txt \
+  --encoder=./sherpa-onnx-streaming-zipformer-en-2023-06-26/encoder-epoch-99-avg-1-chunk-16-left-64.onnx \
+  --decoder=./sherpa-onnx-streaming-zipformer-en-2023-06-26/decoder-epoch-99-avg-1-chunk-16-left-64.onnx \
+  --joiner=./sherpa-onnx-streaming-zipformer-en-2023-06-26/joiner-epoch-99-avg-1-chunk-16-left-64.onnx \
+  --decoding-method=modified_beam_search \
+  --lm=/path/to/lm.onnx \
+  --lm-scale=0.1 \
+  --lodr-fst=/path/to/lodr.fst \
+  --lodr-scale=-0.1 \
+  ./sherpa-onnx-streaming-zipformer-en-2023-06-26/test_wavs/0.wav \
+  ./sherpa-onnx-streaming-zipformer-en-2023-06-26/test_wavs/1.wav \
+  ./sherpa-onnx-streaming-zipformer-en-2023-06-26/test_wavs/8k.wav
+
 (2) Streaming paraformer
 
 curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-streaming-paraformer-bilingual-zh-en.tar.bz2
@@ -186,6 +202,22 @@ def get_args():
         """,
     )
 
+    parser.add_argument(
+        "--lodr-fst",
+        metavar="file",
+        type=str,
+        default="",
+        help="Path to LODR FST model. Used only when --lm is given.",
+    )
+
+    parser.add_argument(
+        "--lodr-scale",
+        metavar="lodr_scale",
+        type=float,
+        default=-0.1,
+        help="LODR scale for rescoring.Used only when --lodr_fst is given.",
+    )
+
     parser.add_argument(
         "--provider",
         type=str,
@@ -320,6 +352,8 @@ def main():
             max_active_paths=args.max_active_paths,
             lm=args.lm,
             lm_scale=args.lm_scale,
+            lodr_fst=args.lodr_fst,
+            lodr_scale=args.lodr_scale,
             hotwords_file=args.hotwords_file,
             hotwords_score=args.hotwords_score,
             modeling_unit=args.modeling_unit,

scripts/apk/generate-vad-asr-apk-script.py

@@ -565,6 +565,38 @@ def get_models():
 
             ls -lh
 
+            popd
+            """,
+        ),
+        Model(
+            model_name="sherpa-onnx-nemo-parakeet_tdt_ctc_110m-en-36000-int8",
+            idx=33,
+            lang="en",
+            lang2="English",
+            short_name="parakeet_tdt_ctc_110m",
+            cmd="""
+            pushd $model_name
+
+            rm -rfv test_wavs
+
+            ls -lh
+
+            popd
+            """,
+        ),
+        Model(
+            model_name="sherpa-onnx-nemo-parakeet-tdt_ctc-0.6b-ja-35000-int8",
+            idx=34,
+            lang="ja",
+            lang2="Japanese",
+            short_name="parakeet-tdt_ctc_0.6b_ja",
+            cmd="""
+            pushd $model_name
+
+            rm -rfv test_wavs
+
+            ls -lh
+
             popd
             """,
         ),

scripts/dotnet/OfflineCanaryModelConfig.cs

@@ -0,0 +1,32 @@
+/// Copyright (c)  2024.5 by 东风破
+
+using System.Runtime.InteropServices;
+
+namespace SherpaOnnx
+{
+    [StructLayout(LayoutKind.Sequential)]
+    public struct OfflineCanaryModelConfig
+    {
+        public OfflineCanaryModelConfig()
+        {
+            Encoder = "";
+            Decoder = "";
+            SrcLang = "en";
+            TgtLang = "en";
+            UsePnc = 1;
+        }
+        [MarshalAs(UnmanagedType.LPStr)]
+        public string Encoder;
+
+        [MarshalAs(UnmanagedType.LPStr)]
+        public string Decoder;
+
+        [MarshalAs(UnmanagedType.LPStr)]
+        public string SrcLang;
+
+        [MarshalAs(UnmanagedType.LPStr)]
+        public string TgtLang;
+
+        public int UsePnc;
+    }
+}

scripts/dotnet/OfflineModelConfig.cs

@@ -28,6 +28,7 @@ namespace SherpaOnnx
             FireRedAsr = new OfflineFireRedAsrModelConfig();
             Dolphin = new OfflineDolphinModelConfig();
             ZipformerCtc = new OfflineZipformerCtcModelConfig();
+            Canary = new OfflineCanaryModelConfig();
         }
         public OfflineTransducerModelConfig Transducer;
         public OfflineParaformerModelConfig Paraformer;
@@ -62,5 +63,6 @@ namespace SherpaOnnx
         public OfflineFireRedAsrModelConfig FireRedAsr;
         public OfflineDolphinModelConfig Dolphin;
         public OfflineZipformerCtcModelConfig ZipformerCtc;
+        public OfflineCanaryModelConfig Canary;
     }
 }