Add HarmonyOS examples for MatchaTTS. (#1678)

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 = '1.10.35';
+export const HAR_VERSION = '1.10.37';
 export const BUILD_MODE_NAME = 'debug';
 export const DEBUG = true;
 export const TARGET_NAME = 'default';

harmony-os/SherpaOnnxHar/sherpa_onnx/Index.ets

@@ -1,11 +1,6 @@
 export { listRawfileDir, readWave, readWaveFromBinary, } from "libsherpa_onnx.so";
 
-export { CircularBuffer,
-  SileroVadConfig,
-  SpeechSegment,
-  Vad,
-  VadConfig,
-} from './src/main/ets/components/Vad';
+export { CircularBuffer, SileroVadConfig, SpeechSegment, Vad, VadConfig, } from './src/main/ets/components/Vad';
 
 
 export { Samples,
@@ -36,7 +31,8 @@ export { OnlineStream,
   OnlineRecognizer,
 } from './src/main/ets/components/StreamingAsr';
 
-export { OfflineTtsVitsModelConfig,
+export { OfflineTtsMatchaModelConfig,
+  OfflineTtsVitsModelConfig,
   OfflineTtsModelConfig,
   OfflineTtsConfig,
   OfflineTts,

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

@@ -17,8 +17,20 @@ export class OfflineTtsVitsModelConfig {
   public lengthScale: number = 1.0;
 }
 
+export class OfflineTtsMatchaModelConfig {
+  public acousticModel: string = '';
+  public vocoder: string = '';
+  public lexicon: string = '';
+  public tokens: string = '';
+  public dataDir: string = '';
+  public dictDir: String = '';
+  public noiseScale: number = 0.667;
+  public lengthScale: number = 1.0;
+}
+
 export class OfflineTtsModelConfig {
   public vits: OfflineTtsVitsModelConfig = new OfflineTtsVitsModelConfig();
+  public matcha: OfflineTtsMatchaModelConfig = new OfflineTtsMatchaModelConfig();
   public numThreads: number = 1;
   public debug: boolean = false;
   public provider: string = 'cpu';

harmony-os/SherpaOnnxTts/entry/src/main/ets/workers/NonStreamingTtsWorker.ets

@@ -73,7 +73,16 @@ function initTts(context: Context): OfflineTts {
   // for details
 
   let modelDir = '';
+
+  // for VITS begin
   let modelName = '';
+  // for VITS end
+
+  // for Matcha begin
+  let acousticModelName = '';
+  let vocoder = '';
+  // for Matcha end
+
   let ruleFsts = '';
   let ruleFars = '';
   let lexicon = '';
@@ -134,15 +143,47 @@ function initTts(context: Context): OfflineTts {
   // dictDir = 'dict';
   // ruleFsts = `date.fst,phone.fst,number.fst`;
 
+  // Example 8
+  // https://github.com/k2-fsa/sherpa-onnx/releases/tag/tts-models
+  // https://k2-fsa.github.io/sherpa/onnx/tts/pretrained_models/matcha.html#matcha-icefall-zh-baker-chinese-1-female-speaker
+  // modelDir = 'matcha-icefall-zh-baker'
+  // acousticModelName = 'model-steps-3.onnx'
+  // vocoder = 'hifigan_v2.onnx'
+  // lexicon = 'lexicon.txt'
+  // dictDir = 'dict';
+  // ruleFsts = `date.fst,phone.fst,number.fst`;
+
+  // Example 9
+  // https://github.com/k2-fsa/sherpa-onnx/releases/tag/tts-models
+  // https://k2-fsa.github.io/sherpa/onnx/tts/pretrained_models/matcha.html#matcha-icefall-en-us-ljspeech-american-english-1-female-speaker
+  // modelDir = 'matcha-icefall-en_US-ljspeech'
+  // acousticModelName = 'model-steps-3.onnx'
+  // vocoder = 'hifigan_v2.onnx'
+  // dataDir = 'espeak-ng-data';
+
   // ============================================================
   // Please don't change the remaining part of this function
   // ============================================================
 
-  if (modelName == '') {
+  if (modelName == '' && acousticModelName == '' && vocoder == '') {
     throw new Error('You are supposed to select a model by changing the code before you run the app');
   }
 
-  modelName = modelDir + '/' + modelName;
+  if (modelName != '' && acousticModelName != '') {
+    throw new Error('Please select either VITS or Matcha, not both');
+  }
+
+  if (acousticModelName != '' && vocoder == '') {
+    throw new Error('Please provider vocoder for matcha tts models');
+  }
+
+  if (modelName != '') {
+    modelName = modelDir + '/' + modelName;
+  }
+
+  if (acousticModelName != '') {
+    acousticModelName = modelDir + '/' + acousticModelName;
+  }
 
   if (ruleFsts != '') {
     let fsts = ruleFsts.split(',')
@@ -186,6 +227,14 @@ function initTts(context: Context): OfflineTts {
   config.model.vits.tokens = tokens;
   config.model.vits.dataDir = dataDir;
   config.model.vits.dictDir = dictDir;
+
+  config.model.matcha.acousticModel = acousticModelName;
+  config.model.matcha.vocoder = vocoder;
+  config.model.matcha.lexicon = lexicon;
+  config.model.matcha.tokens = tokens;
+  config.model.matcha.dataDir = dataDir;
+  config.model.matcha.dictDir = dictDir;
+
   config.model.numThreads = 2;
   config.model.debug = true;
   config.ruleFsts = ruleFsts;

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

@@ -2098,7 +2098,7 @@ SherpaOnnxCreateSpeakerEmbeddingExtractorOHOS(
 }
 
 #if SHERPA_ONNX_ENABLE_TTS == 1
-SherpaOnnxOfflineTts *SherpaOnnxCreateOfflineTtsOHOS(
+const SherpaOnnxOfflineTts *SherpaOnnxCreateOfflineTtsOHOS(
     const SherpaOnnxOfflineTtsConfig *config, NativeResourceManager *mgr) {
   if (!mgr) {
     return SherpaOnnxCreateOfflineTts(config);

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

@@ -1618,7 +1618,7 @@ SherpaOnnxCreateVoiceActivityDetectorOHOS(
     const SherpaOnnxVadModelConfig *config, float buffer_size_in_seconds,
     NativeResourceManager *mgr);
 
-SHERPA_ONNX_API SherpaOnnxOfflineTts *SherpaOnnxCreateOfflineTtsOHOS(
+SHERPA_ONNX_API const SherpaOnnxOfflineTts *SherpaOnnxCreateOfflineTtsOHOS(
     const SherpaOnnxOfflineTtsConfig *config, NativeResourceManager *mgr);
 
 SHERPA_ONNX_API const SherpaOnnxSpeakerEmbeddingExtractor *

sherpa-onnx/csrc/jieba-lexicon.cc

@@ -6,12 +6,23 @@
 
 #include <fstream>
 #include <regex>  // NOLINT
+#include <strstream>
 #include <unordered_set>
 #include <utility>
 
+#if __ANDROID_API__ >= 9
+#include "android/asset_manager.h"
+#include "android/asset_manager_jni.h"
+#endif
+
+#if __OHOS__
+#include "rawfile/raw_file_manager.h"
+#endif
+
 #include "cppjieba/Jieba.hpp"
 #include "sherpa-onnx/csrc/file-utils.h"
 #include "sherpa-onnx/csrc/macros.h"
+#include "sherpa-onnx/csrc/onnx-utils.h"
 #include "sherpa-onnx/csrc/symbol-table.h"
 #include "sherpa-onnx/csrc/text-utils.h"
 
@@ -56,6 +67,39 @@ class JiebaLexicon::Impl {
     }
   }
 
+  template <typename Manager>
+  Impl(Manager *mgr, const std::string &lexicon, const std::string &tokens,
+       const std::string &dict_dir, bool debug)
+      : debug_(debug) {
+    std::string dict = dict_dir + "/jieba.dict.utf8";
+    std::string hmm = dict_dir + "/hmm_model.utf8";
+    std::string user_dict = dict_dir + "/user.dict.utf8";
+    std::string idf = dict_dir + "/idf.utf8";
+    std::string stop_word = dict_dir + "/stop_words.utf8";
+
+    AssertFileExists(dict);
+    AssertFileExists(hmm);
+    AssertFileExists(user_dict);
+    AssertFileExists(idf);
+    AssertFileExists(stop_word);
+
+    jieba_ =
+        std::make_unique<cppjieba::Jieba>(dict, hmm, user_dict, idf, stop_word);
+
+    {
+      auto buf = ReadFile(mgr, tokens);
+      std::istrstream is(buf.data(), buf.size());
+
+      InitTokens(is);
+    }
+
+    {
+      auto buf = ReadFile(mgr, lexicon);
+      std::istrstream is(buf.data(), buf.size());
+      InitLexicon(is);
+    }
+  }
+
   std::vector<TokenIDs> ConvertTextToTokenIds(const std::string &text) const {
     // see
     // https://github.com/Plachtaa/VITS-fast-fine-tuning/blob/main/text/mandarin.py#L244
@@ -279,9 +323,29 @@ JiebaLexicon::JiebaLexicon(const std::string &lexicon,
                            const std::string &dict_dir, bool debug)
     : impl_(std::make_unique<Impl>(lexicon, tokens, dict_dir, debug)) {}
 
+template <typename Manager>
+JiebaLexicon::JiebaLexicon(Manager *mgr, const std::string &lexicon,
+                           const std::string &tokens,
+                           const std::string &dict_dir, bool debug)
+    : impl_(std::make_unique<Impl>(mgr, lexicon, tokens, dict_dir, debug)) {}
+
 std::vector<TokenIDs> JiebaLexicon::ConvertTextToTokenIds(
     const std::string &text, const std::string & /*unused_voice = ""*/) const {
   return impl_->ConvertTextToTokenIds(text);
 }
 
+#if __ANDROID_API__ >= 9
+template JiebaLexicon::JiebaLexicon(AAssetManager *mgr,
+                                    const std::string &lexicon,
+                                    const std::string &tokens,
+                                    const std::string &dict_dir, bool debug);
+#endif
+
+#if __OHOS__
+template JiebaLexicon::JiebaLexicon(NativeResourceManager *mgr,
+                                    const std::string &lexicon,
+                                    const std::string &tokens,
+                                    const std::string &dict_dir, bool debug);
+#endif
+
 }  // namespace sherpa_onnx

sherpa-onnx/csrc/jieba-lexicon.h

@@ -17,9 +17,15 @@ namespace sherpa_onnx {
 class JiebaLexicon : public OfflineTtsFrontend {
  public:
   ~JiebaLexicon() override;
+
   JiebaLexicon(const std::string &lexicon, const std::string &tokens,
                const std::string &dict_dir, bool debug);
 
+  template <typename Manager>
+  JiebaLexicon(Manager *mgr, const std::string &lexicon,
+               const std::string &tokens, const std::string &dict_dir,
+               bool debug);
+
   std::vector<TokenIDs> ConvertTextToTokenIds(
       const std::string &text,
       const std::string &unused_voice = "") const override;

sherpa-onnx/csrc/offline-tts-matcha-impl.h

@@ -327,13 +327,12 @@ class OfflineTtsMatchaImpl : public OfflineTtsImpl {
     // from assets to disk
     //
     // for jieba
-    // we require that you copy tokens.txt, lexicon.txt and dict
-    // from assets to disk
+    // we require that you copy dict from assets to disk
     const auto &meta_data = model_->GetMetaData();
 
     if (meta_data.jieba && !meta_data.has_espeak) {
       frontend_ = std::make_unique<JiebaLexicon>(
-          config_.model.matcha.lexicon, config_.model.matcha.tokens,
+          mgr, config_.model.matcha.lexicon, config_.model.matcha.tokens,
           config_.model.matcha.dict_dir, config_.model.debug);
     } else if (meta_data.has_espeak && !meta_data.jieba) {
       frontend_ = std::make_unique<PiperPhonemizeLexicon>(