Add CXX API for MatchaTTS models (#1676)

@@ -101,6 +101,10 @@ jobs:
 
           ./matcha-tts-zh-c-api
 
+          rm ./matcha-tts-zh-c-api
+          rm -rf matcha-icefall-*
+          rm hifigan_v2.onnx
+
       - name: Test Matcha TTS (en)
         shell: bash
         run: |
@@ -121,6 +125,10 @@ jobs:
 
           ./matcha-tts-en-c-api
 
+          rm ./matcha-tts-en-c-api
+          rm -rf matcha-icefall-*
+          rm hifigan_v2.onnx
+
       - uses: actions/upload-artifact@v4
         with:
           name: matcha-tts-${{ matrix.os }}

@@ -83,6 +83,61 @@ jobs:
             otool -L ./install/lib/libsherpa-onnx-cxx-api.dylib
           fi
 
+      - name: Test Matcha TTS (zh)
+        shell: bash
+        run: |
+          g++ -std=c++17 -o matcha-tts-zh-cxx-api ./cxx-api-examples/matcha-tts-zh-cxx-api.cc \
+            -I ./build/install/include \
+            -L ./build/install/lib/ \
+            -l sherpa-onnx-cxx-api \
+            -l sherpa-onnx-c-api \
+            -l onnxruntime
+
+          curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/matcha-icefall-zh-baker.tar.bz2
+          tar xvf matcha-icefall-zh-baker.tar.bz2
+          rm matcha-icefall-zh-baker.tar.bz2
+
+          curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/vocoder-models/hifigan_v2.onnx
+
+          export LD_LIBRARY_PATH=$PWD/build/install/lib:$LD_LIBRARY_PATH
+          export DYLD_LIBRARY_PATH=$PWD/build/install/lib:$DYLD_LIBRARY_PATH
+
+          ./matcha-tts-zh-cxx-api
+
+          rm -rf matcha-icefall-*
+          rm hifigan_v2.onnx
+          rm matcha-tts-zh-cxx-api
+
+      - name: Test Matcha TTS (en)
+        shell: bash
+        run: |
+          g++ -std=c++17 -o matcha-tts-en-cxx-api ./cxx-api-examples/matcha-tts-en-cxx-api.cc \
+            -I ./build/install/include \
+            -L ./build/install/lib/ \
+            -l sherpa-onnx-cxx-api \
+            -l sherpa-onnx-c-api \
+            -l onnxruntime
+
+          curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/matcha-icefall-en_US-ljspeech.tar.bz2
+          tar xvf matcha-icefall-en_US-ljspeech.tar.bz2
+          rm matcha-icefall-en_US-ljspeech.tar.bz2
+
+          curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/vocoder-models/hifigan_v2.onnx
+
+          export LD_LIBRARY_PATH=$PWD/build/install/lib:$LD_LIBRARY_PATH
+          export DYLD_LIBRARY_PATH=$PWD/build/install/lib:$DYLD_LIBRARY_PATH
+
+          ./matcha-tts-en-cxx-api
+
+          rm matcha-tts-en-cxx-api
+          rm -rf matcha-icefall-*
+          rm hifigan_v2.onnx
+
+      - uses: actions/upload-artifact@v4
+        with:
+          name: matcha-tts-${{ matrix.os }}
+          path: ./generated-matcha-*.wav
+
       - name: Test Moonshine tiny
         shell: bash
         run: |

@@ -60,7 +60,7 @@ int32_t main(int32_t argc, char *argv[]) {
       "Friends fell out often because life was changing so fast. The easiest "
       "thing in the world was to lose touch with someone.";
 
-  SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);
+  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);
   int32_t sid = 0;
   float speed = 1.0;  // larger -> faster in speech speed
 

@@ -60,7 +60,7 @@ int32_t main(int32_t argc, char *argv[]) {
       "某某银行的副行长和一些行政领导表示，他们去过长江和长白山; "
       "经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。";
 
-  SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);
+  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);
   int32_t sid = 0;
   float speed = 1.0;  // larger -> faster in speech speed
 

@@ -229,7 +229,7 @@ int32_t main(int32_t argc, char *argv[]) {
     ShowUsage();
   }
 
-  SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);
+  const SherpaOnnxOfflineTts *tts = SherpaOnnxCreateOfflineTts(&config);
 
   const SherpaOnnxGeneratedAudio *audio =
       SherpaOnnxOfflineTtsGenerate(tts, text, sid, 1.0);

@@ -14,3 +14,11 @@ 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)
+
+if(SHERPA_ONNX_ENABLE_TTS)
+  add_executable(matcha-tts-zh-cxx-api ./matcha-tts-zh-cxx-api.cc)
+  target_link_libraries(matcha-tts-zh-cxx-api sherpa-onnx-cxx-api)
+
+  add_executable(matcha-tts-en-cxx-api ./matcha-tts-en-cxx-api.cc)
+  target_link_libraries(matcha-tts-en-cxx-api sherpa-onnx-cxx-api)
+endif()

+// cxx-api-examples/matcha-tts-en-cxx-api.c
+//
+// Copyright (c)  2025  Xiaomi Corporation
+
+// This file shows how to use sherpa-onnx CXX API
+// for Chinese TTS with MatchaTTS.
+//
+// clang-format off
+/*
+Usage
+
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/matcha-icefall-en_US-ljspeech.tar.bz2
+tar xvf matcha-icefall-en_US-ljspeech.tar.bz2
+rm matcha-icefall-en_US-ljspeech.tar.bz2
+
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/vocoder-models/hifigan_v2.onnx
+
+./matcha-tts-en-cxx-api
+
+ */
+// clang-format on
+
+#include <string>
+
+#include "sherpa-onnx/c-api/cxx-api.h"
+
+static int32_t ProgressCallback(const float *samples, int32_t num_samples,
+                                float progress, void *arg) {
+  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
+  // return 1 to continue generating
+  // return 0 to stop generating
+  return 1;
+}
+
+int32_t main(int32_t argc, char *argv[]) {
+  using namespace sherpa_onnx::cxx;  // NOLINT
+  OfflineTtsConfig config;
+
+  config.model.matcha.acoustic_model =
+      "./matcha-icefall-en_US-ljspeech/model-steps-3.onnx";
+
+  config.model.matcha.vocoder = "./hifigan_v2.onnx";
+
+  config.model.matcha.tokens = "./matcha-icefall-en_US-ljspeech/tokens.txt";
+
+  config.model.matcha.data_dir =
+      "./matcha-icefall-en_US-ljspeech/espeak-ng-data";
+
+  config.model.num_threads = 1;
+
+  // If you don't want to see debug messages, please set it to 0
+  config.model.debug = 1;
+
+  std::string filename = "./generated-matcha-en-cxx.wav";
+  std::string text =
+      "Today as always, men fall into two groups: slaves and free men. Whoever "
+      "does not have two-thirds of his day for himself, is a slave, whatever "
+      "he may be: a statesman, a businessman, an official, or a scholar. "
+      "Friends fell out often because life was changing so fast. The easiest "
+      "thing in the world was to lose touch with someone.";
+
+  auto tts = OfflineTts::Create(config);
+  int32_t sid = 0;
+  float speed = 1.0;  // larger -> faster in speech speed
+
+#if 0
+  // If you don't want to use a callback, then please enable this branch
+  GeneratedAudio audio = tts.Generate(text, sid, speed);
+#else
+  GeneratedAudio audio = tts.Generate(text, sid, speed, ProgressCallback);
+#endif
+
+  WriteWave(filename, {audio.samples, audio.sample_rate});
+
+  fprintf(stderr, "Input text is: %s\n", text.c_str());
+  fprintf(stderr, "Speaker ID is is: %d\n", sid);
+  fprintf(stderr, "Saved to: %s\n", filename.c_str());
+
+  return 0;
+}

+// cxx-api-examples/matcha-tts-zh-cxx-api.c
+//
+// Copyright (c)  2025  Xiaomi Corporation
+
+// This file shows how to use sherpa-onnx CXX API
+// for Chinese TTS with MatchaTTS.
+//
+// clang-format off
+/*
+Usage
+
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/matcha-icefall-zh-baker.tar.bz2
+tar xvf matcha-icefall-zh-baker.tar.bz2
+rm matcha-icefall-zh-baker.tar.bz2
+
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/vocoder-models/hifigan_v2.onnx
+
+./matcha-tts-zh-cxx-api
+
+ */
+// clang-format on
+
+#include <string>
+
+#include "sherpa-onnx/c-api/cxx-api.h"
+
+static int32_t ProgressCallback(const float *samples, int32_t num_samples,
+                                float progress, void *arg) {
+  fprintf(stderr, "Progress: %.3f%%\n", progress * 100);
+  // return 1 to continue generating
+  // return 0 to stop generating
+  return 1;
+}
+
+int32_t main(int32_t argc, char *argv[]) {
+  using namespace sherpa_onnx::cxx;  // NOLINT
+  OfflineTtsConfig config;
+  config.model.matcha.acoustic_model =
+      "./matcha-icefall-zh-baker/model-steps-3.onnx";
+  config.model.matcha.vocoder = "./hifigan_v2.onnx";
+  config.model.matcha.lexicon = "./matcha-icefall-zh-baker/lexicon.txt";
+  config.model.matcha.tokens = "./matcha-icefall-zh-baker/tokens.txt";
+  config.model.matcha.dict_dir = "./matcha-icefall-zh-baker/dict";
+  config.model.num_threads = 1;
+
+  // If you don't want to see debug messages, please set it to 0
+  config.model.debug = 1;
+
+  // clang-format off
+  config.rule_fsts = "./matcha-icefall-zh-baker/phone.fst,./matcha-icefall-zh-baker/date.fst,./matcha-icefall-zh-baker/number.fst";  // NOLINT
+  // clang-format on
+
+  std::string filename = "./generated-matcha-zh-cxx.wav";
+  std::string text =
+      "当夜幕降临，星光点点，伴随着微风拂面，我在静谧中感受着时光的流转，思念如"
+      "涟漪荡漾，梦境如画卷展开，我与自然融为一体，沉静在这片宁静的美丽之中，感"
+      "受着生命的奇迹与温柔."
+      "某某银行的副行长和一些行政领导表示，他们去过长江和长白山; "
+      "经济不断增长。2024年12月31号，拨打110或者18920240511。123456块钱。";
+
+  auto tts = OfflineTts::Create(config);
+  int32_t sid = 0;
+  float speed = 1.0;  // larger -> faster in speech speed
+
+#if 0
+  // If you don't want to use a callback, then please enable this branch
+  GeneratedAudio audio = tts.Generate(text, sid, speed);
+#else
+  GeneratedAudio audio = tts.Generate(text, sid, speed, ProgressCallback);
+#endif
+
+  WriteWave(filename, {audio.samples, audio.sample_rate});
+
+  fprintf(stderr, "Input text is: %s\n", text.c_str());
+  fprintf(stderr, "Speaker ID is is: %d\n", sid);
+  fprintf(stderr, "Saved to: %s\n", filename.c_str());
+
+  return 0;
+}

@@ -1114,7 +1114,7 @@ static sherpa_onnx::OfflineTtsConfig GetOfflineTtsConfig(
   return tts_config;
 }
 
-SherpaOnnxOfflineTts *SherpaOnnxCreateOfflineTts(
+const SherpaOnnxOfflineTts *SherpaOnnxCreateOfflineTts(
     const SherpaOnnxOfflineTtsConfig *config) {
   auto tts_config = GetOfflineTtsConfig(config);
 
@@ -1130,7 +1130,9 @@ SherpaOnnxOfflineTts *SherpaOnnxCreateOfflineTts(
   return tts;
 }
 
-void SherpaOnnxDestroyOfflineTts(SherpaOnnxOfflineTts *tts) { delete tts; }
+void SherpaOnnxDestroyOfflineTts(const SherpaOnnxOfflineTts *tts) {
+  delete tts;
+}
 
 int32_t SherpaOnnxOfflineTtsSampleRate(const SherpaOnnxOfflineTts *tts) {
   return tts->impl->SampleRate();

@@ -950,11 +950,12 @@ SHERPA_ONNX_API typedef struct SherpaOnnxOfflineTts SherpaOnnxOfflineTts;
 
 // Create an instance of offline TTS. The user has to use DestroyOfflineTts()
 // to free the returned pointer to avoid memory leak.
-SHERPA_ONNX_API SherpaOnnxOfflineTts *SherpaOnnxCreateOfflineTts(
+SHERPA_ONNX_API const SherpaOnnxOfflineTts *SherpaOnnxCreateOfflineTts(
     const SherpaOnnxOfflineTtsConfig *config);
 
 // Free the pointer returned by SherpaOnnxCreateOfflineTts()
-SHERPA_ONNX_API void SherpaOnnxDestroyOfflineTts(SherpaOnnxOfflineTts *tts);
+SHERPA_ONNX_API void SherpaOnnxDestroyOfflineTts(
+    const SherpaOnnxOfflineTts *tts);
 
 // Return the sample rate of the current TTS object
 SHERPA_ONNX_API int32_t
@@ -984,7 +985,6 @@ SHERPA_ONNX_API
 const SherpaOnnxGeneratedAudio *
 SherpaOnnxOfflineTtsGenerateWithProgressCallback(
     const SherpaOnnxOfflineTts *tts, const char *text, int32_t sid, float speed,
-
     SherpaOnnxGeneratedAudioProgressCallback callback);
 
 SHERPA_ONNX_API

@@ -24,6 +24,11 @@ Wave ReadWave(const std::string &filename) {
   return ans;
 }
 
+bool WriteWave(const std::string &filename, const Wave &wave) {
+  return SherpaOnnxWriteWave(wave.samples.data(), wave.samples.size(),
+                             wave.sample_rate, filename.c_str());
+}
+
 OnlineStream::OnlineStream(const SherpaOnnxOnlineStream *p)
     : MoveOnly<OnlineStream, SherpaOnnxOnlineStream>(p) {}
 
@@ -311,4 +316,73 @@ OfflineRecognizerResult OfflineRecognizer::GetResult(
   return ans;
 }
 
+OfflineTts OfflineTts::Create(const OfflineTtsConfig &config) {
+  struct SherpaOnnxOfflineTtsConfig c;
+  memset(&c, 0, sizeof(c));
+
+  c.model.vits.model = config.model.vits.model.c_str();
+  c.model.vits.lexicon = config.model.vits.lexicon.c_str();
+  c.model.vits.tokens = config.model.vits.tokens.c_str();
+  c.model.vits.data_dir = config.model.vits.data_dir.c_str();
+  c.model.vits.noise_scale = config.model.vits.noise_scale;
+  c.model.vits.noise_scale_w = config.model.vits.noise_scale_w;
+  c.model.vits.length_scale = config.model.vits.length_scale;
+  c.model.vits.dict_dir = config.model.vits.dict_dir.c_str();
+
+  c.model.matcha.acoustic_model = config.model.matcha.acoustic_model.c_str();
+  c.model.matcha.vocoder = config.model.matcha.vocoder.c_str();
+  c.model.matcha.lexicon = config.model.matcha.lexicon.c_str();
+  c.model.matcha.tokens = config.model.matcha.tokens.c_str();
+  c.model.matcha.data_dir = config.model.matcha.data_dir.c_str();
+  c.model.matcha.noise_scale = config.model.matcha.noise_scale;
+  c.model.matcha.length_scale = config.model.matcha.length_scale;
+  c.model.matcha.dict_dir = config.model.matcha.dict_dir.c_str();
+
+  c.model.num_threads = config.model.num_threads;
+  c.model.debug = config.model.debug;
+  c.model.provider = config.model.provider.c_str();
+
+  c.rule_fsts = config.rule_fsts.c_str();
+  c.max_num_sentences = config.max_num_sentences;
+  c.rule_fars = config.rule_fars.c_str();
+
+  auto p = SherpaOnnxCreateOfflineTts(&c);
+  return OfflineTts(p);
+}
+
+OfflineTts::OfflineTts(const SherpaOnnxOfflineTts *p)
+    : MoveOnly<OfflineTts, SherpaOnnxOfflineTts>(p) {}
+
+void OfflineTts::Destroy(const SherpaOnnxOfflineTts *p) const {
+  SherpaOnnxDestroyOfflineTts(p);
+}
+
+int32_t OfflineTts::SampleRate() const {
+  return SherpaOnnxOfflineTtsSampleRate(p_);
+}
+
+int32_t OfflineTts::NumSpeakers() const {
+  return SherpaOnnxOfflineTtsNumSpeakers(p_);
+}
+
+GeneratedAudio OfflineTts::Generate(const std::string &text,
+                                    int32_t sid /*= 0*/, float speed /*= 1.0*/,
+                                    OfflineTtsCallback callback /*= nullptr*/,
+                                    void *arg /*= nullptr*/) const {
+  const SherpaOnnxGeneratedAudio *audio;
+  if (!callback) {
+    audio = SherpaOnnxOfflineTtsGenerate(p_, text.c_str(), sid, speed);
+  } else {
+    audio = SherpaOnnxOfflineTtsGenerateWithProgressCallbackWithArg(
+        p_, text.c_str(), sid, speed, callback, arg);
+  }
+
+  GeneratedAudio ans;
+  ans.samples = std::vector<float>{audio->samples, audio->samples + audio->n};
+  ans.sample_rate = audio->sample_rate;
+
+  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
+  return ans;
+}
+
 }  // namespace sherpa_onnx::cxx

@@ -97,6 +97,10 @@ struct Wave {
 
 SHERPA_ONNX_API Wave ReadWave(const std::string &filename);
 
+// Return true on success;
+// Return false on failure
+SHERPA_ONNX_API bool WriteWave(const std::string &filename, const Wave &wave);
+
 template <typename Derived, typename T>
 class SHERPA_ONNX_API MoveOnly {
  public:
@@ -307,6 +311,91 @@ class SHERPA_ONNX_API OfflineRecognizer
   explicit OfflineRecognizer(const SherpaOnnxOfflineRecognizer *p);
 };
 
+// ============================================================================
+// Non-streaming TTS
+// ============================================================================
+struct OfflineTtsVitsModelConfig {
+  std::string model;
+  std::string lexicon;
+  std::string tokens;
+  std::string data_dir;
+  std::string dict_dir;
+
+  float noise_scale = 0.667;
+  float noise_scale_w = 0.8;
+  float length_scale = 1.0;  // < 1, faster in speed; > 1, slower in speed
+};
+
+struct OfflineTtsMatchaModelConfig {
+  std::string acoustic_model;
+  std::string vocoder;
+  std::string lexicon;
+  std::string tokens;
+  std::string data_dir;
+  std::string dict_dir;
+
+  float noise_scale = 0.667;
+  float length_scale = 1.0;  // < 1, faster in speed; > 1, slower in speed
+};
+
+struct OfflineTtsModelConfig {
+  OfflineTtsVitsModelConfig vits;
+  OfflineTtsMatchaModelConfig matcha;
+  int32_t num_threads = 1;
+  bool debug = false;
+  std::string provider = "cpu";
+};
+
+struct OfflineTtsConfig {
+  OfflineTtsModelConfig model;
+  std::string rule_fsts;
+  std::string rule_fars;
+  int32_t max_num_sentences = 1;
+};
+
+struct GeneratedAudio {
+  std::vector<float> samples;  // in the range [-1, 1]
+  int32_t sample_rate;
+};
+
+// Return 1 to continue generating
+// Return 0 to stop generating
+using OfflineTtsCallback = int32_t (*)(const float *samples,
+                                       int32_t num_samples, float progress,
+                                       void *arg);
+
+class SHERPA_ONNX_API OfflineTts
+    : public MoveOnly<OfflineTts, SherpaOnnxOfflineTts> {
+ public:
+  static OfflineTts Create(const OfflineTtsConfig &config);
+
+  void Destroy(const SherpaOnnxOfflineTts *p) const;
+
+  // Return the sample rate of the generated audio
+  int32_t SampleRate() const;
+
+  // Number of supported speakers.
+  // If it supports only a single speaker, then it return 0 or 1.
+  int32_t NumSpeakers() const;
+
+  // @param text A string containing words separated by spaces
+  // @param sid Speaker ID. Used only for multi-speaker models, e.g., models
+  //            trained using the VCTK dataset. It is not used for
+  //            single-speaker models, e.g., models trained using the ljspeech
+  //            dataset.
+  // @param speed The speed for the generated speech. E.g., 2 means 2x faster.
+  // @param callback If not NULL, it is called whenever config.max_num_sentences
+  //                 sentences have been processed. The callback is called in
+  //                 the current thread.
+  GeneratedAudio Generate(const std::string &text, int32_t sid = 0,
+                          float speed = 1.0,
+                          OfflineTtsCallback callback = nullptr,
+                          void *arg = nullptr) const;
+
+ private:
+  explicit OfflineTts(const SherpaOnnxOfflineTts *p);
+};
+
 }  // namespace sherpa_onnx::cxx
 
 #endif  // SHERPA_ONNX_C_API_CXX_API_H_