Add C and CXX API for Dolphin CTC models (#2088)

.github/workflows/c-api.yaml

@@ -79,6 +79,36 @@ jobs:
             otool -L ./install/lib/libsherpa-onnx-c-api.dylib
           fi
 
+      - name: Test Dolphin CTC
+        shell: bash
+        run: |
+          name=dolphin-ctc-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 ./$name
+          fi
+
+          curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02
+          tar xvf sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02.tar.bz2
+          rm sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02.tar.bz2
+
+          export LD_LIBRARY_PATH=$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-dolphin-base-ctc-multi-lang-int8-2025-04-02
+
       - name: Test speech enhancement (GTCRN)
         shell: bash
         run: |

.github/workflows/cxx-api.yaml

@@ -81,6 +81,38 @@ jobs:
             otool -L ./install/lib/libsherpa-onnx-cxx-api.dylib
           fi
 
+      - name: Test Dolphin CTC
+        shell: bash
+        run: |
+          name=dolphin-ctc-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 $name
+
+          export LD_LIBRARY_PATH=$PWD/build/install/lib:$LD_LIBRARY_PATH
+          export DYLD_LIBRARY_PATH=$PWD/build/install/lib:$DYLD_LIBRARY_PATH
+
+          if [[ ${{ matrix.os }} == ubuntu-latest || ${{ matrix.os }} == ubuntu-22.04-arm ]]; then
+            ldd ./$name
+            echo "----"
+            readelf -d ./$name
+          fi
+
+          curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02
+          tar xvf sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02.tar.bz2
+          rm sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02.tar.bz2
+
+          ./$name
+
+          rm -rf sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02
+
+          rm $name
+
       - name: Test VAD
         shell: bash
         run: |

c-api-examples/dolphin-ctc-c-api.c

@@ -0,0 +1,74 @@
+// c-api-examples/dolphin-ctc-c-api.c
+//
+// Copyright (c)  2025  Xiaomi Corporation
+
+//
+// This file demonstrates how to use Dolphin CTC model with sherpa-onnx's C API.
+// clang-format off
+//
+// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02
+// tar xvf sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02.tar.bz2
+// rm sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02.tar.bz2
+//
+// clang-format on
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "sherpa-onnx/c-api/c-api.h"
+
+int32_t main() {
+  // clang-format off
+  const char *wav_filename = "./sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02/test_wavs/0.wav";
+  const char *model_filename = "./sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02/model.int8.onnx";
+  const char *tokens_filename = "./sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02/tokens.txt";
+  // clang-format on
+
+  const SherpaOnnxWave *wave = SherpaOnnxReadWave(wav_filename);
+  if (wave == NULL) {
+    fprintf(stderr, "Failed to read %s\n", wav_filename);
+    return -1;
+  }
+
+  SherpaOnnxOfflineModelConfig offline_model_config;
+  memset(&offline_model_config, 0, sizeof(offline_model_config));
+  offline_model_config.debug = 1;
+  offline_model_config.num_threads = 1;
+  offline_model_config.provider = "cpu";
+  offline_model_config.tokens = tokens_filename;
+  offline_model_config.dolphin.model = model_filename;
+
+  // 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: %s\n", result->text);
+
+  SherpaOnnxDestroyOfflineRecognizerResult(result);
+  SherpaOnnxDestroyOfflineStream(stream);
+  SherpaOnnxDestroyOfflineRecognizer(recognizer);
+  SherpaOnnxFreeWave(wave);
+
+  return 0;
+}

cxx-api-examples/CMakeLists.txt

@@ -24,6 +24,9 @@ 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(dolphin-ctc-cxx-api ./dolphin-ctc-cxx-api.cc)
+target_link_libraries(dolphin-ctc-cxx-api sherpa-onnx-cxx-api)
+
 add_executable(vad-cxx-api ./vad-cxx-api.cc)
 target_link_libraries(vad-cxx-api sherpa-onnx-cxx-api)
 

cxx-api-examples/dolphin-ctc-cxx-api.cc

@@ -0,0 +1,76 @@
+// cxx-api-examples/dolphin-ctc-cxx-api.cc
+// Copyright (c)  2025  Xiaomi Corporation
+
+//
+// This file demonstrates how to use Dolphini CTC model with sherpa-onnx's C++
+// API.
+//
+// clang-format off
+//
+// wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02
+// tar xvf sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02.tar.bz2
+// rm sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02.tar.bz2
+//
+// clang-format on
+
+#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;
+
+  // clang-format off
+  config.model_config.dolphin.model = "./sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02/model.int8.onnx";
+  config.model_config.tokens = "./sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02/tokens.txt";
+
+  std::string wave_filename = "./sherpa-onnx-dolphin-base-ctc-multi-lang-int8-2025-04-02/test_wavs/0.wav";
+  // clang-format on
+
+  config.model_config.num_threads = 1;
+
+  std::cout << "Loading model\n";
+  OfflineRecognizer recongizer = OfflineRecognizer::Create(config);
+  if (!recongizer.Get()) {
+    std::cerr << "Please check your config\n";
+    return -1;
+  }
+  std::cout << "Loading model done\n";
+
+  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 = recongizer.CreateStream();
+  stream.AcceptWaveform(wave.sample_rate, wave.samples.data(),
+                        wave.samples.size());
+
+  recongizer.Decode(&stream);
+
+  OfflineRecognizerResult result = recongizer.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: " << 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);
+
+  return 0;
+}

scripts/apk/generate-asr-2pass-apk-script.py

@@ -333,7 +333,6 @@ def get_1st_models():
             rm -f bpe.model
 
             rm -rf test_wavs
-            rm README.md
 
             ls -lh
 
@@ -354,7 +353,6 @@ def get_1st_models():
             rm -f bpe.model
 
             rm -rf test_wavs
-            rm README.md
 
             ls -lh
 

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

@@ -277,7 +277,6 @@ def get_models():
             rm -f bpe.model
 
             rm -rf test_wavs
-            rm README.md
 
             ls -lh
 
@@ -298,7 +297,6 @@ def get_models():
             rm -f bpe.model
 
             rm -rf test_wavs
-            rm README.md
 
             ls -lh
 

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

@@ -448,7 +448,7 @@ def get_models():
             idx=25,
             lang="multi_lang",
             lang2="multi_lang",
-            short_name="multi_lang",
+            short_name="dolphin_base_ctc",
             cmd="""
             pushd $model_name
 

sherpa-onnx/c-api/c-api.cc

@@ -467,6 +467,9 @@ static sherpa_onnx::OfflineRecognizerConfig GetOfflineRecognizerConfig(
   recognizer_config.model_config.fire_red_asr.decoder =
       SHERPA_ONNX_OR(config->model_config.fire_red_asr.decoder, "");
 
+  recognizer_config.model_config.dolphin.model =
+      SHERPA_ONNX_OR(config->model_config.dolphin.model, "");
+
   recognizer_config.lm_config.model =
       SHERPA_ONNX_OR(config->lm_config.model, "");
   recognizer_config.lm_config.scale =

sherpa-onnx/c-api/c-api.h

@@ -416,6 +416,10 @@ SHERPA_ONNX_API typedef struct SherpaOnnxOfflineSenseVoiceModelConfig {
   int32_t use_itn;
 } SherpaOnnxOfflineSenseVoiceModelConfig;
 
+SHERPA_ONNX_API typedef struct SherpaOnnxOfflineDolphinModelConfig {
+  const char *model;
+} SherpaOnnxOfflineDolphinModelConfig;
+
 SHERPA_ONNX_API typedef struct SherpaOnnxOfflineModelConfig {
   SherpaOnnxOfflineTransducerModelConfig transducer;
   SherpaOnnxOfflineParaformerModelConfig paraformer;
@@ -438,6 +442,7 @@ SHERPA_ONNX_API typedef struct SherpaOnnxOfflineModelConfig {
   SherpaOnnxOfflineSenseVoiceModelConfig sense_voice;
   SherpaOnnxOfflineMoonshineModelConfig moonshine;
   SherpaOnnxOfflineFireRedAsrModelConfig fire_red_asr;
+  SherpaOnnxOfflineDolphinModelConfig dolphin;
 } SherpaOnnxOfflineModelConfig;
 
 SHERPA_ONNX_API typedef struct SherpaOnnxOfflineRecognizerConfig {

sherpa-onnx/c-api/cxx-api.cc

@@ -246,6 +246,8 @@ OfflineRecognizer OfflineRecognizer::Create(
   c.model_config.fire_red_asr.decoder =
       config.model_config.fire_red_asr.decoder.c_str();
 
+  c.model_config.dolphin.model = config.model_config.dolphin.model.c_str();
+
   c.lm_config.model = config.lm_config.model.c_str();
   c.lm_config.scale = config.lm_config.scale;
 

sherpa-onnx/c-api/cxx-api.h

@@ -229,6 +229,10 @@ struct SHERPA_ONNX_API OfflineSenseVoiceModelConfig {
   bool use_itn = false;
 };
 
+struct SHERPA_ONNX_API OfflineDolphinModelConfig {
+  std::string model;
+};
+
 struct SHERPA_ONNX_API OfflineMoonshineModelConfig {
   std::string preprocessor;
   std::string encoder;
@@ -254,6 +258,7 @@ struct SHERPA_ONNX_API OfflineModelConfig {
   OfflineSenseVoiceModelConfig sense_voice;
   OfflineMoonshineModelConfig moonshine;
   OfflineFireRedAsrModelConfig fire_red_asr;
+  OfflineDolphinModelConfig dolphin;
 };
 
 struct SHERPA_ONNX_API OfflineLMConfig {