Add PengChengStarling models to sherpa-onnx (#1835)

.github/workflows/export-peng-cheng-starling.yaml

@@ -0,0 +1,123 @@
+name: export-peng-cheng-starling-to-onnx
+
+on:
+  push:
+    branches:
+      - export-peng-cheng-starling-2
+
+  workflow_dispatch:
+
+concurrency:
+  group: export-peng-cheng-starling-to-onnx-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  export-peng-cheng-starling-to-onnx:
+    if: github.repository_owner == 'k2-fsa' || github.repository_owner == 'csukuangfj'
+    name: export peng cheng starling ${{ matrix.version }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-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 "numpy<=1.26.4" onnx==1.16.0 onnxruntime==1.17.1
+
+      - name: Run
+        shell: bash
+        run: |
+          cd scripts/peng-cheng-starling
+          ./run.sh
+          python3 ./quantize_models.py
+
+          ls -lh
+          rm encoder-epoch-75-avg-11-chunk-16-left-128.onnx
+          rm joiner-epoch-75-avg-11-chunk-16-left-128.onnx
+          echo "----"
+          ls -lh
+
+
+      - name: Collect results ${{ matrix.version }}
+        shell: bash
+        run: |
+          src=scripts/peng-cheng-starling
+          d=sherpa-onnx-streaming-zipformer-ar_en_id_ja_ru_th_vi_zh-2025-02-10
+          mkdir $d
+
+          mv -v $src/*.onnx $d
+          cp -v $src/README.md $d
+          cp -v $src/bpe.model $d
+          cp -av $src/test_wavs $d
+
+          ls -lh $d/
+          tar cjfv $d.tar.bz2 $d
+
+          ls -lh $d.tar.bz2
+
+      - name: Publish to huggingface ${{ matrix.version }}
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+        uses: nick-fields/retry@v3
+        with:
+          max_attempts: 20
+          timeout_seconds: 200
+          shell: bash
+          command: |
+            git config --global user.email "csukuangfj@gmail.com"
+            git config --global user.name "Fangjun Kuang"
+
+            rm -rf huggingface
+            export GIT_LFS_SKIP_SMUDGE=1
+            export GIT_CLONE_PROTECTION_ACTIVE=false
+
+            src=sherpa-onnx-streaming-zipformer-ar_en_id_ja_ru_th_vi_zh-2025-02-10
+
+            git clone https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$src huggingface
+            cd huggingface
+            rm -rf ./*
+            git fetch
+            git pull
+
+            cp -av ../$src/* ./
+
+            git lfs track "*.onnx"
+            git add .
+
+            ls -lh
+
+            git status
+
+            git commit -m "add models"
+            git push https://csukuangfj:$HF_TOKEN@huggingface.co/csukuangfj/$src main || true
+
+      - name: Release
+        if: github.repository_owner == 'csukuangfj'
+        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
+
+      - name: Release
+        if: github.repository_owner == 'k2-fsa'
+        uses: svenstaro/upload-release-action@v2
+        with:
+          file_glob: true
+          file: ./*.tar.bz2
+          overwrite: true
+          tag: asr-models

scripts/matcha-tts/README.md

@@ -1,6 +1,6 @@
 # Introduction
 
-This folder contains script for adding meta data to tts models
+This folder contains scripts for adding meta data to tts models
 from https://github.com/shivammehta25/Matcha-TTS
 
 Note: If you use icefall to train a MatchaTTS model, you don't need this folder.

scripts/peng-cheng-starling/.gitignore

@@ -0,0 +1,3 @@
+bpe.model
+*.wav
+*.onnx

scripts/peng-cheng-starling/README.md

@@ -0,0 +1,4 @@
+# Introduction
+
+This folder contains scripts for files from
+https://github.com/yangb05/PengChengStarling

scripts/peng-cheng-starling/quantize_models.py

@@ -0,0 +1,22 @@
+#!/usr/bin/env python3
+from onnxruntime.quantization import QuantType, quantize_dynamic
+from pathlib import Path
+
+
+def main():
+    suffix = "epoch-75-avg-11-chunk-16-left-128"
+
+    for m in ["encoder", "joiner"]:
+        if Path(f"{m}-{suffix}.int8.onnx").is_file():
+            continue
+
+        quantize_dynamic(
+            model_input=f"./{m}-{suffix}.onnx",
+            model_output=f"./{m}-{suffix}.int8.onnx",
+            op_types_to_quantize=["MatMul"],
+            weight_type=QuantType.QInt8,
+        )
+
+
+if __name__ == "__main__":
+    main()

scripts/peng-cheng-starling/run.sh

@@ -0,0 +1,36 @@
+#!/usr/bin/env bash
+
+set -ex
+
+if [ ! -f bpe.model ]; then
+  curl -SL -O https://huggingface.co/stdo/PengChengStarling/resolve/main/bpe.model
+fi
+
+if [ ! -f tokens.txt ]; then
+  curl -SL -O https://huggingface.co/stdo/PengChengStarling/resolve/main/tokens.txt
+fi
+
+if [ ! -f decoder-epoch-75-avg-11-chunk-16-left-128.onnx ]; then
+  curl -SL -O https://huggingface.co/stdo/PengChengStarling/resolve/main/decoder-epoch-75-avg-11-chunk-16-left-128.onnx
+fi
+
+if [ ! -f encoder-epoch-75-avg-11-chunk-16-left-128.onnx ]; then
+  curl -SL -O https://huggingface.co/stdo/PengChengStarling/resolve/main/encoder-epoch-75-avg-11-chunk-16-left-128.onnx
+fi
+
+if [ ! -f joiner-epoch-75-avg-11-chunk-16-left-128.onnx ]; then
+  curl -SL -O https://huggingface.co/stdo/PengChengStarling/resolve/main/joiner-epoch-75-avg-11-chunk-16-left-128.onnx
+fi
+
+mkdir -p test_wavs
+if [ ! -f test_wavs/zh.wav ]; then
+  curl -SL --output test_wavs/zh.wav https://huggingface.co/marcoyang/sherpa-ncnn-streaming-zipformer-zh-14M-2023-02-23/resolve/main/test_wavs/0.wav
+fi
+
+if [ ! -f test_wavs/en.wav ]; then
+  curl -SL --output test_wavs/en.wav https://huggingface.co/csukuangfj/sherpa-onnx-streaming-zipformer-en-2023-02-21/resolve/main/test_wavs/0.wav
+fi
+
+if [ ! -f test_wavs/ja.wav ]; then
+  curl -SL --output test_wavs/ja.wav https://huggingface.co/csukuangfj/reazonspeech-k2-v2/resolve/main/test_wavs/5.wav
+fi

sherpa-onnx/csrc/offline-tts.h

@@ -32,7 +32,7 @@ struct OfflineTtsConfig {
   // If you set it to -1, then we process all sentences in a single batch.
   int32_t max_num_sentences = 1;
 
-  // A silence interval containing audio samples with value close to 0.
+  // A silence interval contains audio samples with value close to 0.
   //
   // the duration of the new interval is old_duration * silence_scale.
   float silence_scale = 0.2;