Dart, flutter, go, c-api blinding and example (#3202)

c-api-examples/pocket-tts-en-c-api.c

@@ -49,6 +49,10 @@ int32_t main(int32_t argc, char *argv[]) {
       "./sherpa-onnx-pocket-tts-int8-2026-01-26/vocab.json";
   config.model.pocket.token_scores_json =
       "./sherpa-onnx-pocket-tts-int8-2026-01-26/token_scores.json";
+  // Voice embedding cache capacity (default: 50)
+  // Increase this if you have many different reference audios to avoid
+  // recomputing voice embeddings
+  config.model.pocket.voice_embedding_cache_capacity = 50;
 
   config.model.num_threads = 2;
 
@@ -82,7 +86,10 @@ int32_t main(int32_t argc, char *argv[]) {
   cfg.reference_audio = wave->samples;
   cfg.reference_audio_len = wave->num_samples;
   cfg.reference_sample_rate = wave->sample_rate;
-  cfg.extra = "{\"max_reference_audio_len\": 10.0}";
+  // Extra parameters passed as JSON string
+  // - max_reference_audio_len: maximum length of reference audio in seconds
+  // - seed: random seed for reproducibility (optional, -1 for random)
+  cfg.extra = "{\"max_reference_audio_len\": 10.0, \"seed\": 42}";
 
 #if 0
   // If you don't want to use a callback, then please enable this branch

dart-api-examples/tts/bin/pocket-en.dart

@@ -19,7 +19,17 @@ void main(List<String> arguments) async {
     ..addOption('token-scores-json', help: 'Path to the token_scores.json file')
     ..addOption('reference-audio', help: 'Path to reference audio (wav)')
     ..addOption('text', help: 'Text to generate TTS for')
-    ..addOption('output-wav', help: 'Filename to save the generated audio');
+    ..addOption('output-wav', help: 'Filename to save the generated audio')
+    ..addOption(
+      'voice-embedding-cache-capacity',
+      help: 'Voice embedding cache capacity (default: 50)',
+      defaultsTo: '50',
+    )
+    ..addOption(
+      'seed',
+      help: 'Random seed for reproducibility (default: -1, random)',
+      defaultsTo: '-1',
+    );
 
   final res = parser.parse(arguments);
 
@@ -47,6 +57,10 @@ void main(List<String> arguments) async {
   final referenceAudioPath = res['reference-audio'] as String;
   final text = res['text'] as String;
   final outputWav = res['output-wav'] as String;
+  final voiceEmbeddingCacheCapacity = int.parse(
+    res['voice-embedding-cache-capacity'] as String,
+  );
+  final seed = int.parse(res['seed'] as String);
 
   // ---------------- Pocket model config ----------------
   final pocket = sherpa_onnx.OfflineTtsPocketModelConfig(
@@ -57,6 +71,7 @@ void main(List<String> arguments) async {
     textConditioner: textConditioner,
     vocabJson: vocabJson,
     tokenScoresJson: tokenScoresJson,
+    voiceEmbeddingCacheCapacity: voiceEmbeddingCacheCapacity,
   );
 
   final modelConfig = sherpa_onnx.OfflineTtsModelConfig(
@@ -80,7 +95,7 @@ void main(List<String> arguments) async {
     speed: 1.0,
     referenceAudio: wave.samples,
     referenceSampleRate: wave.sampleRate,
-    extra: {"max_reference_audio_len": 12},
+    extra: {"max_reference_audio_len": 12, if (seed >= 0) "seed": seed},
   );
 
   // If you don't want to use a callback

flutter/sherpa_onnx/lib/src/sherpa_onnx_bindings.dart

@@ -243,6 +243,9 @@ final class SherpaOnnxOfflineTtsPocketModelConfig extends Struct {
   external Pointer<Utf8> textConditioner;
   external Pointer<Utf8> vocabJson;
   external Pointer<Utf8> tokenScoresJson;
+
+  @Int32()
+  external int voiceEmbeddingCacheCapacity;
 }
 
 final class SherpaOnnxOfflineTtsModelConfig extends Struct {

flutter/sherpa_onnx/lib/src/tts.dart

@@ -352,6 +352,7 @@ class OfflineTtsPocketModelConfig {
     this.textConditioner = '',
     this.vocabJson = '',
     this.tokenScoresJson = '',
+    this.voiceEmbeddingCacheCapacity = 50,
   });
 
   factory OfflineTtsPocketModelConfig.fromJson(Map<String, dynamic> json) {
@@ -363,6 +364,8 @@ class OfflineTtsPocketModelConfig {
       textConditioner: json['textConditioner'] as String? ?? '',
       vocabJson: json['vocabJson'] as String? ?? '',
       tokenScoresJson: json['tokenScoresJson'] as String? ?? '',
+      voiceEmbeddingCacheCapacity:
+          json['voiceEmbeddingCacheCapacity'] as int? ?? 50,
     );
   }
 
@@ -374,11 +377,12 @@ class OfflineTtsPocketModelConfig {
     'textConditioner': textConditioner,
     'vocabJson': vocabJson,
     'tokenScoresJson': tokenScoresJson,
+    'voiceEmbeddingCacheCapacity': voiceEmbeddingCacheCapacity,
   };
 
   @override
   String toString() {
-    return 'OfflineTtsPocketModelConfig(lmFlow: $lmFlow, lmMain: $lmMain, encoder: $encoder, decoder: $decoder, textConditioner: $textConditioner, vocabJson: $vocabJson, tokenScoresJson: $tokenScoresJson)';
+    return 'OfflineTtsPocketModelConfig(lmFlow: $lmFlow, lmMain: $lmMain, encoder: $encoder, decoder: $decoder, textConditioner: $textConditioner, vocabJson: $vocabJson, tokenScoresJson: $tokenScoresJson, voiceEmbeddingCacheCapacity: $voiceEmbeddingCacheCapacity)';
   }
 
   final String lmFlow;
@@ -388,6 +392,7 @@ class OfflineTtsPocketModelConfig {
   final String textConditioner;
   final String vocabJson;
   final String tokenScoresJson;
+  final int voiceEmbeddingCacheCapacity;
 }
 
 class OfflineTtsModelConfig {
@@ -569,6 +574,8 @@ class OfflineTts {
     c.ref.model.pocket.vocabJson = config.model.pocket.vocabJson.toNativeUtf8();
     c.ref.model.pocket.tokenScoresJson = config.model.pocket.tokenScoresJson
         .toNativeUtf8();
+    c.ref.model.pocket.voiceEmbeddingCacheCapacity =
+        config.model.pocket.voiceEmbeddingCacheCapacity;
 
     c.ref.model.numThreads = config.model.numThreads;
     c.ref.model.debug = config.model.debug ? 1 : 0;

go-api-examples/non-streaming-funasr-nano-decode-files/main.go

@@ -20,6 +20,8 @@ func main() {
 	config.ModelConfig.FunAsrNano.LLM = "./sherpa-onnx-funasr-nano-int8-2025-12-30/llm.int8.onnx"
 	config.ModelConfig.FunAsrNano.Embedding = "./sherpa-onnx-funasr-nano-int8-2025-12-30/embedding.int8.onnx"
 	config.ModelConfig.FunAsrNano.Tokenizer = "./sherpa-onnx-funasr-nano-int8-2025-12-30/Qwen3-0.6B"
+	// Seed for reproducibility (default: 42)
+	config.ModelConfig.FunAsrNano.Seed = 42
 
 	config.ModelConfig.Tokens = ""
 

go-api-examples/zero-shot-pocket-tts-play/main.go

@@ -94,6 +94,8 @@ func main() {
 
 	var referenceAudio string = "./sherpa-onnx-pocket-tts-int8-2026-01-26/test_wavs/bria.wav"
 	var outputFilename string = "./generated.wav"
+	var voiceEmbeddingCacheCapacity int = 50
+	var seed int = -1
 
 	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,
@@ -102,6 +104,8 @@ whatever he may be: a statesman, a businessman, an official, or a scholar.`
 	flag.StringVar(&referenceAudio, "reference-audio", referenceAudio, "Path to the reference audio")
 	flag.StringVar(&text, "text", text, "Text to be synthesized")
 	flag.StringVar(&outputFilename, "output-filename", outputFilename, "File to save the generated audio")
+	flag.IntVar(&voiceEmbeddingCacheCapacity, "voice-embedding-cache-capacity", voiceEmbeddingCacheCapacity, "Voice embedding cache capacity (default: 50)")
+	flag.IntVar(&seed, "seed", seed, "Random seed for reproducibility (default: -1, random)")
 	flag.Parse()
 
 	// ---------------- config ----------------
@@ -121,6 +125,7 @@ whatever he may be: a statesman, a businessman, an official, or a scholar.`
 		"./sherpa-onnx-pocket-tts-int8-2026-01-26/vocab.json"
 	config.Model.Pocket.TokenScoresJson =
 		"./sherpa-onnx-pocket-tts-int8-2026-01-26/token_scores.json"
+	config.Model.Pocket.VoiceEmbeddingCacheCapacity = voiceEmbeddingCacheCapacity
 
 	config.Model.NumThreads = 2
 	config.Model.Debug = 0
@@ -142,10 +147,16 @@ whatever he may be: a statesman, a businessman, an official, or a scholar.`
 	cfg.ReferenceAudio = wave.Samples
 	cfg.ReferenceSampleRate = wave.SampleRate
 
-	cfg.Extra = json.RawMessage(`{
-  "max_reference_audio_len": 10,
-	"temperature": 0.7
-	}`)
+	// Build extra config with optional seed
+	extraMap := map[string]interface{}{
+		"max_reference_audio_len": 10,
+		"temperature":             0.7,
+	}
+	if seed >= 0 {
+		extraMap["seed"] = seed
+	}
+	extraBytes, _ := json.Marshal(extraMap)
+	cfg.Extra = json.RawMessage(extraBytes)
 
 	log.Println("Start generating")
 

go-api-examples/zero-shot-pocket-tts/main.go

@@ -13,6 +13,8 @@ func main() {
 
 	var referenceAudio string = "./sherpa-onnx-pocket-tts-int8-2026-01-26/test_wavs/bria.wav"
 	var outputFilename string = "./generated.wav"
+	var voiceEmbeddingCacheCapacity int = 50
+	var seed int = -1
 
 	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,
@@ -21,6 +23,8 @@ whatever he may be: a statesman, a businessman, an official, or a scholar.`
 	flag.StringVar(&referenceAudio, "reference-audio", referenceAudio, "Path to the reference audio")
 	flag.StringVar(&text, "text", text, "Text to be synthesized")
 	flag.StringVar(&outputFilename, "output-filename", outputFilename, "File to save the generated audio")
+	flag.IntVar(&voiceEmbeddingCacheCapacity, "voice-embedding-cache-capacity", voiceEmbeddingCacheCapacity, "Voice embedding cache capacity (default: 50)")
+	flag.IntVar(&seed, "seed", seed, "Random seed for reproducibility (default: -1, random)")
 	flag.Parse()
 
 	// ---------------- config ----------------
@@ -40,6 +44,7 @@ whatever he may be: a statesman, a businessman, an official, or a scholar.`
 		"./sherpa-onnx-pocket-tts-int8-2026-01-26/vocab.json"
 	config.Model.Pocket.TokenScoresJson =
 		"./sherpa-onnx-pocket-tts-int8-2026-01-26/token_scores.json"
+	config.Model.Pocket.VoiceEmbeddingCacheCapacity = voiceEmbeddingCacheCapacity
 
 	config.Model.NumThreads = 2
 	config.Model.Debug = 0
@@ -61,10 +66,16 @@ whatever he may be: a statesman, a businessman, an official, or a scholar.`
 	cfg.ReferenceAudio = wave.Samples
 	cfg.ReferenceSampleRate = wave.SampleRate
 
-	cfg.Extra = json.RawMessage(`{
-  "max_reference_audio_len": 10,
-	"temperature": 0.7
-	}`)
+	// Build extra config with optional seed
+	extraMap := map[string]interface{}{
+		"max_reference_audio_len": 10,
+		"temperature":             0.7,
+	}
+	if seed >= 0 {
+		extraMap["seed"] = seed
+	}
+	extraBytes, _ := json.Marshal(extraMap)
+	cfg.Extra = json.RawMessage(extraBytes)
 
 	log.Println("Start generating")