csukuangfj · Dec 11, 2025 · Dec 11, 2025
diff --git a/flutter/sherpa_onnx/lib/src/sherpa_onnx_bindings.dart b/flutter/sherpa_onnx/lib/src/sherpa_onnx_bindings.dart
@@ -220,7 +220,7 @@ final class SherpaOnnxOfflineTtsZipVoiceModelConfig extends Struct {
   external Pointer<Utf8> flowMatchingModel;
   external Pointer<Utf8> vocoder;
   external Pointer<Utf8> dataDir;
-  external Pointer<Utf8> pinyinDict;
+  external Pointer<Utf8> lexicon;
 
   @Float()
   external double featScale;

diff --git a/flutter/sherpa_onnx/lib/src/tts.dart b/flutter/sherpa_onnx/lib/src/tts.dart
@@ -199,7 +199,7 @@ class OfflineTtsZipVoiceModelConfig {
     this.flowMatchingModel = '',
     this.vocoder = '',
     this.dataDir = '',
-    this.pinyinDict = '',
+    this.lexicon = '',
     this.featScale = 0.1,
     this.tShift = 0.5,
     this.targetRms = 0.1,
@@ -213,7 +213,7 @@ class OfflineTtsZipVoiceModelConfig {
       flowMatchingModel: json['flowMatchingModel'] as String? ?? '',
       vocoder: json['vocoder'] as String? ?? '',
       dataDir: json['dataDir'] as String? ?? '',
-      pinyinDict: json['pinyinDict'] as String? ?? '',
+      lexicon: json['lexicon'] as String? ?? '',
       featScale: (json['featScale'] as num?)?.toDouble() ?? 0.1,
       tShift: (json['tShift'] as num?)?.toDouble() ?? 0.5,
       targetRms: (json['targetRms'] as num?)?.toDouble() ?? 0.1,
@@ -223,7 +223,7 @@ class OfflineTtsZipVoiceModelConfig {
 
   @override
   String toString() {
-    return 'OfflineTtsZipVoiceModelConfig(tokens: $tokens, textModel: $textModel, flowMatchingModel: $flowMatchingModel, vocoder: $vocoder, dataDir: $dataDir, pinyinDict: $pinyinDict, featScale: $featScale, tShift: $tShift, targetRms: $targetRms, guidanceScale: $guidanceScale)';
+    return 'OfflineTtsZipVoiceModelConfig(tokens: $tokens, textModel: $textModel, flowMatchingModel: $flowMatchingModel, vocoder: $vocoder, dataDir: $dataDir, lexicon: $lexicon, featScale: $featScale, tShift: $tShift, targetRms: $targetRms, guidanceScale: $guidanceScale)';
   }
 
   Map<String, dynamic> toJson() => {
@@ -232,7 +232,7 @@ class OfflineTtsZipVoiceModelConfig {
         'flowMatchingModel': flowMatchingModel,
         'vocoder': vocoder,
         'dataDir': dataDir,
-        'pinyinDict': pinyinDict,
+        'lexicon': lexicon,
         'featScale': featScale,
         'tShift': tShift,
         'targetRms': targetRms,
@@ -244,7 +244,7 @@ class OfflineTtsZipVoiceModelConfig {
   final String flowMatchingModel;
   final String vocoder;
   final String dataDir;
-  final String pinyinDict;
+  final String lexicon;
   final double featScale;
   final double tShift;
   final double targetRms;
@@ -406,7 +406,7 @@ class OfflineTts {
     c.ref.model.zipvoice.flowMatchingModel = config.model.zipvoice.flowMatchingModel.toNativeUtf8();
     c.ref.model.zipvoice.vocoder = config.model.zipvoice.vocoder.toNativeUtf8();
     c.ref.model.zipvoice.dataDir = config.model.zipvoice.dataDir.toNativeUtf8();
-    c.ref.model.zipvoice.pinyinDict = config.model.zipvoice.pinyinDict.toNativeUtf8();
+    c.ref.model.zipvoice.lexicon = config.model.zipvoice.lexicon.toNativeUtf8();
     c.ref.model.zipvoice.featScale = config.model.zipvoice.featScale;
     c.ref.model.zipvoice.tShift = config.model.zipvoice.tShift;
     c.ref.model.zipvoice.targetRms = config.model.zipvoice.targetRms;
@@ -427,7 +427,7 @@ class OfflineTts {
     calloc.free(c.ref.ruleFsts);
     calloc.free(c.ref.model.provider);
 
-    calloc.free(c.ref.model.zipvoice.pinyinDict);
+    calloc.free(c.ref.model.zipvoice.lexicon);
     calloc.free(c.ref.model.zipvoice.dataDir);
     calloc.free(c.ref.model.zipvoice.vocoder);
     calloc.free(c.ref.model.zipvoice.flowMatchingModel);

diff --git a/python-api-examples/offline-zeroshot-tts.py b/python-api-examples/offline-zeroshot-tts.py
@@ -10,21 +10,23 @@
 
 Example (zipvoice)
 
-wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/sherpa-onnx-zipvoice-distill-zh-en-emilia.tar.bz2
-tar xf sherpa-onnx-zipvoice-distill-zh-en-emilia.tar.bz2
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/sherpa-onnx-zipvoice-distill-int8-zh-en-emilia.tar.bz2
+tar xf sherpa-onnx-zipvoice-distill-int8-zh-en-emilia.tar.bz2
+
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/vocoder-models/vocos_24khz.onnx
 
 python3 ./python-api-examples/offline-zeroshot-tts.py \
-  --zipvoice-flow-matching-model sherpa-onnx-zipvoice-distill-zh-en-emilia/fm_decoder.onnx \
-  --zipvoice-text-model sherpa-onnx-zipvoice-distill-zh-en-emilia/text_encoder.onnx \
-  --zipvoice-data-dir sherpa-onnx-zipvoice-distill-zh-en-emilia/espeak-ng-data \
-  --zipvoice-pinyin-dict sherpa-onnx-zipvoice-distill-zh-en-emilia/pinyin.raw \
-  --zipvoice-tokens sherpa-onnx-zipvoice-distill-zh-en-emilia/tokens.txt \
-  --zipvoice-vocoder sherpa-onnx-zipvoice-distill-zh-en-emilia/vocos_24khz.onnx \
-  --prompt-audio sherpa-onnx-zipvoice-distill-zh-en-emilia/prompt.wav \
+  --zipvoice-flow-matching-model sherpa-onnx-zipvoice-distill-int8-zh-en-emilia/decoder.int8.onnx \
+  --zipvoice-text-model sherpa-onnx-zipvoice-distill-int8-zh-en-emilia/encoder.int8.onnx \
+  --zipvoice-data-dir sherpa-onnx-zipvoice-distill-int8-zh-en-emilia/espeak-ng-data \
+  --zipvoice-lexicon sherpa-onnx-zipvoice-distill-int8-zh-en-emilia/lexicon.txt \
+  --zipvoice-tokens sherpa-onnx-zipvoice-distill-int8-zh-en-emilia/tokens.txt \
+  --zipvoice-vocoder vocos_24khz.onnx \
+  --prompt-audio sherpa-onnx-zipvoice-distill-int8-zh-en-emilia/test_wavs/leijun-1.wav \
   --zipvoice-num-steps 4 \
   --num-threads 4 \
-  --prompt-text "周日被我射熄火了，所以今天是周一。" \
-  "我是中国人民的儿子，我爱我的祖国。我得祖国是一个伟大的国家，拥有五千年的文明史。"
+  --prompt-text "那还是三十六年前, 一九八七年. 我呢考上了武汉大学的计算机系." \
+  "小米的价值观是真诚, 热爱. 真诚，就是不欺人也不自欺. 热爱, 就是全心投入并享受其中."
 """
 
 import argparse
@@ -68,10 +70,10 @@ def add_zipvoice_args(parser):
     )
 
     parser.add_argument(
-        "--zipvoice-pinyin-dict",
+        "--zipvoice-lexicon",
         type=str,
         default="",
-        help="Path to the pinyin dictionary.",
+        help="Path to the lexicon.txt",
     )
 
     parser.add_argument(
@@ -236,7 +238,7 @@ def main():
                 text_model=args.zipvoice_text_model,
                 flow_matching_model=args.zipvoice_flow_matching_model,
                 data_dir=args.zipvoice_data_dir,
-                pinyin_dict=args.zipvoice_pinyin_dict,
+                lexicon=args.zipvoice_lexicon,
                 vocoder=args.zipvoice_vocoder,
                 feat_scale=args.zipvoice_feat_scale,
                 t_shift=args.zipvoice_t_shift,
@@ -268,9 +270,7 @@ def main():
     end = time.time()
 
     if len(audio.samples) == 0:
-        print(
-            "Error in generating audios. Please read previous error messages."
-        )
+        print("Error in generating audios. Please read previous error messages.")
         return
 
     elapsed_seconds = end - start
@@ -287,9 +287,7 @@ def main():
     print(f"The text is '{args.text}'")
     print(f"Elapsed seconds: {elapsed_seconds:.3f}")
     print(f"Audio duration in seconds: {audio_duration:.3f}")
-    print(
-        f"RTF: {elapsed_seconds:.3f}/{audio_duration:.3f} = {real_time_factor:.3f}"
-    )
+    print(f"RTF: {elapsed_seconds:.3f}/{audio_duration:.3f} = {real_time_factor:.3f}")
 
 
 if __name__ == "__main__":

diff --git a/scripts/go/sherpa_onnx.go b/scripts/go/sherpa_onnx.go
@@ -484,7 +484,7 @@ type OfflineModelConfig struct {
 	ZipformerCtc OfflineZipformerCtcModelConfig
 	Canary       OfflineCanaryModelConfig
 	WenetCtc     OfflineWenetCtcModelConfig
-	Omnilingual     OfflineOmnilingualAsrCtcModelConfig
+	Omnilingual  OfflineOmnilingualAsrCtcModelConfig
 	Tokens       string // Path to tokens.txt
 
 	// Number of threads to use for neural network computation
@@ -964,7 +964,7 @@ type OfflineTtsZipvoiceModelConfig struct {
 	TextModel         string // Path to text encoder (e.g. text_encoder.onnx)
 	FlowMatchingModel string // Path to flow-matching decoder (e.g. fm_decoder.onnx)
 	DataDir           string // Path to espeak-ng-data
-	PinyinDict        string // Path to pinyin.raw (needed for zh)
+	Lexicon           string // Path to lexicon.txt (needed for zh)
 	Vocoder           string // Path to vocoder (e.g. vocos_24khz.onnx)
 
 	FeatScale     float32 // Feature scale
@@ -1148,8 +1148,8 @@ func NewOfflineTts(config *OfflineTtsConfig) *OfflineTts {
 	c.model.zipvoice.data_dir = C.CString(config.Model.Zipvoice.DataDir)
 	defer C.free(unsafe.Pointer(c.model.zipvoice.data_dir))
 
-	c.model.zipvoice.pinyin_dict = C.CString(config.Model.Zipvoice.PinyinDict)
-	defer C.free(unsafe.Pointer(c.model.zipvoice.pinyin_dict))
+	c.model.zipvoice.lexicon = C.CString(config.Model.Zipvoice.Lexicon)
+	defer C.free(unsafe.Pointer(c.model.zipvoice.lexicon))
 
 	c.model.zipvoice.feat_scale = C.float(config.Model.Zipvoice.FeatScale)
 	c.model.zipvoice.t_shift = C.float(config.Model.Zipvoice.TShift)

diff --git a/sherpa-onnx/c-api/c-api.cc b/sherpa-onnx/c-api/c-api.cc
@@ -703,9 +703,11 @@ const SherpaOnnxOfflineRecognizerResult *SherpaOnnxGetOfflineStreamResult(
       r->durations = nullptr;
     }
 
-    if (!result.ys_log_probs.empty() && result.ys_log_probs.size() == r->count) {
+    if (!result.ys_log_probs.empty() &&
+        result.ys_log_probs.size() == r->count) {
       r->ys_log_probs = new float[r->count];
-      std::copy(result.ys_log_probs.begin(), result.ys_log_probs.end(), r->ys_log_probs);
+      std::copy(result.ys_log_probs.begin(), result.ys_log_probs.end(),
+                r->ys_log_probs);
     } else {
       r->ys_log_probs = nullptr;
     }
@@ -1248,8 +1250,8 @@ static sherpa_onnx::OfflineTtsConfig GetOfflineTtsConfig(
       SHERPA_ONNX_OR(config->model.zipvoice.vocoder, "");
   tts_config.model.zipvoice.data_dir =
       SHERPA_ONNX_OR(config->model.zipvoice.data_dir, "");
-  tts_config.model.zipvoice.pinyin_dict =
-      SHERPA_ONNX_OR(config->model.zipvoice.pinyin_dict, "");
+  tts_config.model.zipvoice.lexicon =
+      SHERPA_ONNX_OR(config->model.zipvoice.lexicon, "");
   tts_config.model.zipvoice.feat_scale =
       SHERPA_ONNX_OR(config->model.zipvoice.feat_scale, 0.1f);
   tts_config.model.zipvoice.t_shift =

diff --git a/sherpa-onnx/c-api/c-api.h b/sherpa-onnx/c-api/c-api.h
@@ -1072,7 +1072,7 @@ SHERPA_ONNX_API typedef struct SherpaOnnxOfflineTtsZipvoiceModelConfig {
   const char *flow_matching_model;
   const char *vocoder;
   const char *data_dir;
-  const char *pinyin_dict;
+  const char *lexicon;
   float feat_scale;
   float t_shift;
   float target_rms;

diff --git a/sherpa-onnx/c-api/cxx-api.cc b/sherpa-onnx/c-api/cxx-api.cc
@@ -419,7 +419,7 @@ OfflineTts OfflineTts::Create(const OfflineTtsConfig &config) {
       config.model.zipvoice.flow_matching_model.c_str();
   c.model.zipvoice.vocoder = config.model.zipvoice.vocoder.c_str();
   c.model.zipvoice.data_dir = config.model.zipvoice.data_dir.c_str();
-  c.model.zipvoice.pinyin_dict = config.model.zipvoice.pinyin_dict.c_str();
+  c.model.zipvoice.lexicon = config.model.zipvoice.lexicon.c_str();
   c.model.zipvoice.feat_scale = config.model.zipvoice.feat_scale;
   c.model.zipvoice.t_shift = config.model.zipvoice.t_shift;
   c.model.zipvoice.target_rms = config.model.zipvoice.target_rms;

diff --git a/sherpa-onnx/c-api/cxx-api.h b/sherpa-onnx/c-api/cxx-api.h
@@ -432,7 +432,7 @@ struct OfflineTtsZipvoiceModelConfig {
   std::string flow_matching_model;
   std::string vocoder;
   std::string data_dir;
-  std::string pinyin_dict;
+  std::string lexicon;
 
   float feat_scale = 0.1;
   float t_shift = 0.5;

diff --git a/sherpa-onnx/csrc/offline-tts-zipvoice-model-config.cc b/sherpa-onnx/csrc/offline-tts-zipvoice-model-config.cc
@@ -18,9 +18,6 @@ void OfflineTtsZipvoiceModelConfig::Register(ParseOptions *po) {
   po->Register("zipvoice-data-dir", &data_dir,
                "Path to the directory containing dict for espeak-ng.");
   po->Register("zipvoice-lexicon", &lexicon, "Path to lexicon.txt for Chinese");
-  po->Register("zipvoice-pinyin-dict", &pinyin_dict,
-               "Path to the pinyin dictionary for cppinyin (i.e converting "
-               "Chinese into phones).");
   po->Register("zipvoice-text-model", &text_model,
                "Path to zipvoice text model");
   po->Register("zipvoice-flow-matching-model", &flow_matching_model,

diff --git a/sherpa-onnx/csrc/offline-tts-zipvoice-model-config.h b/sherpa-onnx/csrc/offline-tts-zipvoice-model-config.h
@@ -21,9 +21,6 @@ struct OfflineTtsZipvoiceModelConfig {
   std::string data_dir;
   std::string lexicon;
 
-  // Used for converting Chinese characters to pinyin
-  std::string pinyin_dict;
-
   float feat_scale = 0.1;
   float t_shift = 0.5;
   float target_rms = 0.1;
@@ -35,15 +32,14 @@ struct OfflineTtsZipvoiceModelConfig {
       const std::string &tokens, const std::string &text_model,
       const std::string &flow_matching_model, const std::string &vocoder,
       const std::string &data_dir, const std::string &lexicon,
-      const std::string &pinyin_dict, float feat_scale = 0.1,
-      float t_shift = 0.5, float target_rms = 0.1, float guidance_scale = 1.0)
+      float feat_scale = 0.1, float t_shift = 0.5, float target_rms = 0.1,
+      float guidance_scale = 1.0)
       : tokens(tokens),
         text_model(text_model),
         flow_matching_model(flow_matching_model),
         vocoder(vocoder),
         data_dir(data_dir),
         lexicon(lexicon),
-        pinyin_dict(pinyin_dict),
         feat_scale(feat_scale),
         t_shift(t_shift),
         target_rms(target_rms),

diff --git a/sherpa-onnx/python/csrc/offline-tts-zipvoice-model-config.cc b/sherpa-onnx/python/csrc/offline-tts-zipvoice-model-config.cc
@@ -21,15 +21,15 @@ void PybindOfflineTtsZipvoiceModelConfig(py::module *m) {
                     float, float>(),
            py::arg("tokens"), py::arg("text_model"),
            py::arg("flow_matching_model"), py::arg("vocoder"),
-           py::arg("data_dir") = "", py::arg("pinyin_dict") = "",
+           py::arg("data_dir") = "", py::arg("lexicon") = "",
            py::arg("feat_scale") = 0.1, py::arg("t_shift") = 0.5,
            py::arg("target_rms") = 0.1, py::arg("guidance_scale") = 1.0)
       .def_readwrite("tokens", &PyClass::tokens)
       .def_readwrite("text_model", &PyClass::text_model)
       .def_readwrite("flow_matching_model", &PyClass::flow_matching_model)
       .def_readwrite("vocoder", &PyClass::vocoder)
       .def_readwrite("data_dir", &PyClass::data_dir)
-      .def_readwrite("pinyin_dict", &PyClass::pinyin_dict)
+      .def_readwrite("lexicon", &PyClass::lexicon)
       .def_readwrite("feat_scale", &PyClass::feat_scale)
       .def_readwrite("t_shift", &PyClass::t_shift)
       .def_readwrite("target_rms", &PyClass::target_rms)

diff --git a/swift-api-examples/SherpaOnnx.swift b/swift-api-examples/SherpaOnnx.swift
@@ -933,7 +933,7 @@ func sherpaOnnxOfflineTtsZipvoiceModelConfig(
   flowMatchingModel: String = "",
   vocoder: String = "",
   dataDir: String = "",
-  pinyinDict: String = "",
+  lexicon: String = "",
   featScale: Float = 0.1,
   tShift: Float = 0.5,
   targetRms: Float = 0.1,
@@ -945,7 +945,7 @@ func sherpaOnnxOfflineTtsZipvoiceModelConfig(
     flow_matching_model: toCPointer(flowMatchingModel),
     vocoder: toCPointer(vocoder),
     data_dir: toCPointer(dataDir),
-    pinyin_dict: toCPointer(pinyinDict),
+    lexicon: toCPointer(lexicon),
     feat_scale: featScale,
     t_shift: tShift,
     target_rms: targetRms,

diff --git a/wasm/tts/sherpa-onnx-tts.js b/wasm/tts/sherpa-onnx-tts.js
@@ -269,10 +269,10 @@ function initSherpaOnnxOfflineTtsZipVoiceModelConfig(config, Module) {
       Module.lengthBytesUTF8(config.flowMatchingModel || '') + 1;
   const vocoderLen = Module.lengthBytesUTF8(config.vocoder || '') + 1;
   const dataDirLen = Module.lengthBytesUTF8(config.dataDir || '') + 1;
-  const pinyinDictLen = Module.lengthBytesUTF8(config.pinyinDict || '') + 1;
+  const lexiconLen = Module.lengthBytesUTF8(config.lexicon || '') + 1;
 
   const n = tokensLen + textModelLen + flowMatchingModelLen + vocoderLen +
-      dataDirLen + pinyinDictLen;
+      dataDirLen + lexiconLen;
 
   const buffer = Module._malloc(n);
 
@@ -296,8 +296,8 @@ function initSherpaOnnxOfflineTtsZipVoiceModelConfig(config, Module) {
   Module.stringToUTF8(config.dataDir || '', buffer + offset, dataDirLen);
   offset += dataDirLen;
 
-  Module.stringToUTF8(config.pinyinDict || '', buffer + offset, pinyinDictLen);
-  offset += pinyinDictLen;
+  Module.stringToUTF8(config.lexicon || '', buffer + offset, lexiconLen);
+  offset += lexiconLen;
 
   offset = 0;
   Module.setValue(ptr, buffer + offset, 'i8*');
@@ -316,7 +316,7 @@ function initSherpaOnnxOfflineTtsZipVoiceModelConfig(config, Module) {
   offset += dataDirLen;
 
   Module.setValue(ptr + 20, buffer + offset, 'i8*');
-  offset += pinyinDictLen;
+  offset += lexiconLen;
 
   Module.setValue(ptr + 24, config.featScale || 0.1, 'float');
   Module.setValue(ptr + 28, config.tShift || 0.5, 'float');
@@ -381,7 +381,7 @@ function initSherpaOnnxOfflineTtsModelConfig(config, Module) {
       flowMatchingModel: '',
       vocoder: '',
       dataDir: '',
-      pinyinDict: '',
+      lexicon: '',
       featScale: 0.1,
       tShift: 0.5,
       targetRMS: 0.1,

diff --git a/wasm/tts/sherpa-onnx-wasm-main-tts.cc b/wasm/tts/sherpa-onnx-wasm-main-tts.cc
@@ -79,7 +79,7 @@ void MyPrint(SherpaOnnxOfflineTtsConfig *tts_config) {
   fprintf(stdout, "flow_matching_model: %s\n", zipvoice->flow_matching_model);
   fprintf(stdout, "vocoder: %s\n", zipvoice->vocoder);
   fprintf(stdout, "data_dir: %s\n", zipvoice->data_dir);
-  fprintf(stdout, "pinyin_dict: %s\n", zipvoice->pinyin_dict);
+  fprintf(stdout, "lexicon: %s\n", zipvoice->lexicon);
   fprintf(stdout, "feat scale: %.3f\n", zipvoice->feat_scale);
   fprintf(stdout, "t_shift: %.3f\n", zipvoice->t_shift);
   fprintf(stdout, "target_rms: %.3f\n", zipvoice->target_rms);