react-native-sherpa-onnx 0.3.5 → 0.3.7

package/src/stt/index.ts

@@ -40,7 +40,7 @@ function normalizeSttResult(raw: {
  *
  * @param modelPath - Model path configuration (asset, file, or auto)
  * @param options - Optional preferInt8 and modelType (default: auto)
- * @returns Object with success, detectedModels (array of { type, modelDir }), and modelType (primary detected type)
+ * @returns Object with success, detectedModels (array of { type, modelDir }), modelType (primary detected type), optional error when success is false, and optionally isHardwareSpecificUnsupported
  * @example
  * ```typescript
  * const path = { type: 'asset' as const, path: 'models/sherpa-onnx-whisper-tiny-en' };
@@ -55,15 +55,33 @@ export async function detectSttModel(
   options?: { preferInt8?: boolean; modelType?: STTModelType }
 ): Promise<{
   success: boolean;
+  /** Native validation/detect failure. */
+  error?: string;
   detectedModels: Array<{ type: string; modelDir: string }>;
   modelType?: string;
+  isHardwareSpecificUnsupported?: boolean;
 }> {
   const resolvedPath = await resolveModelPath(modelPath);
-  return SherpaOnnx.detectSttModel(
+  const raw = await SherpaOnnx.detectSttModel(
     resolvedPath,
     options?.preferInt8,
     options?.modelType
   );
+  const err =
+    typeof (raw as { error?: unknown }).error === 'string'
+      ? String((raw as { error: string }).error).trim()
+      : '';
+  return {
+    success: raw.success,
+    ...(err.length > 0 ? { error: err } : {}),
+    ...(raw.isHardwareSpecificUnsupported === true
+      ? { isHardwareSpecificUnsupported: true }
+      : {}),
+    detectedModels: raw.detectedModels ?? [],
+    ...(raw.modelType != null && raw.modelType !== ''
+      ? { modelType: raw.modelType }
+      : {}),
+  };
 }
 
 /**

package/src/stt/streaming.ts

@@ -82,6 +82,7 @@ function flattenInitOptionsForNative(options: StreamingSttInitOptions): {
   provider?: string;
   ruleFsts?: string;
   ruleFars?: string;
+  dither?: number;
   blankPenalty?: number;
   debug?: boolean;
   rule1MustContainNonSilence?: boolean;
@@ -107,6 +108,7 @@ function flattenInitOptionsForNative(options: StreamingSttInitOptions): {
     provider: options.provider,
     ruleFsts: options.ruleFsts,
     ruleFars: options.ruleFars,
+    dither: options.dither,
     blankPenalty: options.blankPenalty,
     debug: options.debug,
     rule1MustContainNonSilence: ep?.rule1?.mustContainNonSilence,
@@ -200,6 +202,7 @@ export async function createStreamingSTT(
   if (flat.provider !== undefined) nativeOptions.provider = flat.provider;
   if (flat.ruleFsts !== undefined) nativeOptions.ruleFsts = flat.ruleFsts;
   if (flat.ruleFars !== undefined) nativeOptions.ruleFars = flat.ruleFars;
+  if (flat.dither !== undefined) nativeOptions.dither = flat.dither;
   if (flat.blankPenalty !== undefined)
     nativeOptions.blankPenalty = flat.blankPenalty;
   if (flat.debug !== undefined) nativeOptions.debug = flat.debug;

package/src/stt/streamingTypes.ts

@@ -75,6 +75,11 @@ export interface StreamingSttInitOptions {
   ruleFsts?: string;
   /** Path(s) to rule FARs for ITN. */
   ruleFars?: string;
+  /**
+   * Feature extraction dither. **Android:** applied natively. **iOS:** ignored (C/CXX API has no
+   * `dither` on `FeatureConfig`); library default applies.
+   */
+  dither?: number;
   /** Blank penalty. */
   blankPenalty?: number;
   /** Enable debug logging. Default: false. */

package/src/stt/types.ts

@@ -228,7 +228,9 @@ export interface STTInitializeOptions {
   ruleFars?: string;
 
   /**
-   * Dither for feature extraction (Kotlin FeatureConfig.dither). Default 0.
+   * Dither for feature extraction (Kotlin `FeatureConfig.dither`). Default: no dither.
+   * **Android:** applied natively. **iOS:** ignored — the bundled sherpa-onnx C/CXX API does not
+   * expose this field; the native default is used.
    */
   dither?: number;
 

package/src/tts/index.ts

@@ -86,7 +86,7 @@ function flattenTtsModelOptionsForNative(
  *
  * @param modelPath - Model path configuration (asset, file, or auto)
  * @param options - Optional modelType (default: 'auto')
- * @returns Object with success, detectedModels (array of { type, modelDir }), modelType (primary detected type), and optionally lexiconLanguageCandidates (language ids for multi-lang Kokoro/Kitten)
+ * @returns Object with success, detectedModels (array of { type, modelDir }), modelType (primary detected type), optional error when success is false, and optionally lexiconLanguageCandidates (language ids for multi-lang Kokoro/Kitten)
  * @example
  * ```typescript
  * const result = await detectTtsModel({ type: 'asset', path: 'models/vits-piper-en' });
@@ -101,13 +101,31 @@ export async function detectTtsModel(
   options?: { modelType?: TTSModelType }
 ): Promise<{
   success: boolean;
+  /** Native validation/detect failure (e.g. missing lexicon for Zipvoice). */
+  error?: string;
   detectedModels: Array<{ type: string; modelDir: string }>;
   modelType?: string;
   /** Language ids from detected lexicon files ("default" for lexicon.txt, or e.g. "us-en", "zh" from lexicon-us-en.txt, lexicon-zh.txt). Present for Kokoro/Kitten; use for language selection UI. */
   lexiconLanguageCandidates?: string[];
 }> {
   const resolvedPath = await resolveModelPath(modelPath);
-  return SherpaOnnx.detectTtsModel(resolvedPath, options?.modelType);
+  const raw = await SherpaOnnx.detectTtsModel(resolvedPath, options?.modelType);
+  const err =
+    typeof (raw as { error?: unknown }).error === 'string'
+      ? String((raw as { error: string }).error).trim()
+      : '';
+  return {
+    success: raw.success,
+    ...(err.length > 0 ? { error: err } : {}),
+    detectedModels: raw.detectedModels ?? [],
+    ...(raw.modelType != null && raw.modelType !== ''
+      ? { modelType: raw.modelType }
+      : {}),
+    ...(raw.lexiconLanguageCandidates != null &&
+    raw.lexiconLanguageCandidates.length > 0
+      ? { lexiconLanguageCandidates: raw.lexiconLanguageCandidates }
+      : {}),
+  };
 }
 
 /**
@@ -124,6 +142,16 @@ function toNativeTtsOptions(
   if (options.silenceScale !== undefined)
     out.silenceScale = options.silenceScale;
   if (options.referenceAudio != null) {
+    const sr = options.referenceAudio.sampleRate;
+    if (
+      typeof __DEV__ !== 'undefined' &&
+      __DEV__ &&
+      (!Number.isFinite(sr) || sr <= 0)
+    ) {
+      console.warn(
+        '[react-native-sherpa-onnx] TTS referenceAudio.sampleRate must be > 0 for voice cloning (Zipvoice/Pocket).'
+      );
+    }
     out.referenceAudio = options.referenceAudio.samples;
     out.referenceSampleRate = options.referenceAudio.sampleRate;
   }

package/src/tts/streaming.ts

@@ -90,6 +90,16 @@ function toNativeTtsOptions(
   if (options.silenceScale !== undefined)
     out.silenceScale = options.silenceScale;
   if (options.referenceAudio != null) {
+    const sr = options.referenceAudio.sampleRate;
+    if (
+      typeof __DEV__ !== 'undefined' &&
+      __DEV__ &&
+      (!Number.isFinite(sr) || sr <= 0)
+    ) {
+      console.warn(
+        '[react-native-sherpa-onnx] TTS referenceAudio.sampleRate must be > 0 for voice cloning (Zipvoice/Pocket).'
+      );
+    }
     out.referenceAudio = options.referenceAudio.samples;
     out.referenceSampleRate = options.referenceAudio.sampleRate;
   }

package/src/tts/types.ts

@@ -202,15 +202,16 @@ export interface TtsGenerationOptions {
   silenceScale?: number;
 
   /**
-   * Reference audio for voice cloning (Kotlin GenerationConfig).
-   * In the Kotlin/RN stack, only Pocket TTS uses this; other model types (vits, matcha, kokoro, kitten) ignore it.
-   * Mono float samples in [-1, 1] and sample rate in Hz.
+   * Reference audio for voice cloning (native GenerationConfig / Zipvoice prompt).
+   * **Native (iOS & Android):** Requires non-empty samples and `sampleRate > 0`. Used for **Zipvoice** (cloning) and **Pocket** (Mimi encoder).
+   * Other model types (vits, matcha, kokoro, kitten) are **rejected** if reference audio is passed.
+   * Mono float samples in [-1, 1].
    */
   referenceAudio?: { samples: number[]; sampleRate: number };
 
   /**
-   * Transcript text of the reference audio (Kotlin GenerationConfig.referenceText).
-   * Required for Pocket TTS when referenceAudio is provided; ignored by other model types.
+   * Transcript of the reference utterance for **Zipvoice** voice cloning (prompt text); **required** when cloning with Zipvoice (non-empty after trim).
+   * **Pocket:** not read by sherpa-onnx native code; optional, e.g. for app metadata only.
    */
   referenceText?: string;
 

package/src/utils.ts

@@ -1,6 +1,7 @@
 import { Platform } from 'react-native';
 import type { ModelPathConfig } from './types';
 import SherpaOnnx from './NativeSherpaOnnx';
+import { resolveActualModelDir } from './download';
 
 /**
  * Utility functions for model path handling
@@ -73,13 +74,33 @@ export function autoModelPath(path: string): ModelPathConfig {
  * This handles different path types (asset, file, auto) and returns
  * a platform-specific absolute path that can be used by native code.
  *
+ * For type 'file', the path is normalized so that when the given path is an
+ * install directory (e.g. with .ready and manifest.json and one model subdir),
+ * the returned path is the subdirectory that actually contains the .onnx files.
+ * This allows apps that build paths as baseDir/modelId to work without change.
+ *
  * @param config - Model path configuration
  * @returns Promise resolving to absolute path usable by native code
  */
 export async function resolveModelPath(
   config: ModelPathConfig
 ): Promise<string> {
-  return SherpaOnnx.resolveModelPath(config);
+  const path = await SherpaOnnx.resolveModelPath(config);
+  if (config.type === 'file') {
+    const resolved = await resolveActualModelDir(path);
+    // Diagnostic: log so we can tell if /usr/share/espeak-ng-data is due to our path or sherpa-onnx fallback.
+    if (__DEV__) {
+      console.log(
+        '[SherpaOnnx] resolveModelPath(file): native path=',
+        path,
+        resolved !== path
+          ? `resolvedActualModelDir=> ${resolved}`
+          : '(unchanged)'
+      );
+    }
+    return resolved;
+  }
+  return path;
 }
 
 /**

package/third_party/libarchive_prebuilt/ANDROID_RELEASE_TAG

@@ -1 +1 @@
-libarchive-android-v3.8.5-1
+libarchive-android-v3.8.5-2

package/third_party/libarchive_prebuilt/IOS_RELEASE_TAG

@@ -1 +1 @@
-libarchive-ios-v3.8.5-1
+libarchive-ios-v3.8.5-2

package/third_party/sherpa-onnx-prebuilt/ANDROID_RELEASE_TAG

@@ -1 +1 @@
-sherpa-onnx-android-v1.12.28
+sherpa-onnx-android-v1.12.31-1

package/third_party/sherpa-onnx-prebuilt/IOS_RELEASE_TAG

@@ -1 +1 @@
-framework-v1.12.28
+sherpa-onnx-ios-v1.12.31-1

package/android/src/main/cpp/jni/tts/sherpa-onnx-tts-zipvoice-jni.cpp

@@ -1,301 +0,0 @@
-/**
- * sherpa-onnx-tts-zipvoice-jni.cpp
- *
- * Purpose: JNI for Zipvoice TTS using the sherpa-onnx C-API (OfflineTtsZipvoiceModelConfig). The
- * Kotlin TTS API does not expose Zipvoice config, so this native layer is used for Zipvoice-only flows.
- */
-#include <jni.h>
-#include <cstring>
-#include <android/log.h>
-
-#include "sherpa-onnx/c-api/c-api.h"
-
-#define LOG_TAG "ZipvoiceTtsJni"
-#define LOGI(...) __android_log_print(ANDROID_LOG_INFO, LOG_TAG, __VA_ARGS__)
-#define LOGE(...) __android_log_print(ANDROID_LOG_ERROR, LOG_TAG, __VA_ARGS__)
-
-namespace {
-
-// Helper: get a non-null C string from a jstring (returns "" for null).
-struct JStringGuard {
-  JNIEnv* env;
-  jstring jstr;
-  const char* cstr;
-
-  JStringGuard(JNIEnv* e, jstring s) : env(e), jstr(s), cstr(nullptr) {
-    if (s) cstr = env->GetStringUTFChars(s, nullptr);
-  }
-  ~JStringGuard() {
-    if (cstr) env->ReleaseStringUTFChars(jstr, cstr);
-  }
-  const char* get() const { return cstr ? cstr : ""; }
-};
-
-// Build a Java float[] + int pair as Object[] { float[], Integer } for returning generated audio.
-jobjectArray buildAudioResult(JNIEnv* env, const float* samples, int32_t n, int32_t sampleRate) {
-  jclass objClass = env->FindClass("java/lang/Object");
-  if (!objClass) return nullptr;
-
-  jobjectArray result = env->NewObjectArray(2, objClass, nullptr);
-  if (!result) {
-    env->DeleteLocalRef(objClass);
-    return nullptr;
-  }
-
-  // Element 0: float[] samples
-  jfloatArray jsamples = env->NewFloatArray(n);
-  if (jsamples && n > 0) {
-    env->SetFloatArrayRegion(jsamples, 0, n, samples);
-  }
-  env->SetObjectArrayElement(result, 0, jsamples);
-  if (jsamples) env->DeleteLocalRef(jsamples);
-
-  // Element 1: Integer sampleRate
-  jclass intClass = env->FindClass("java/lang/Integer");
-  jmethodID intValueOf = env->GetStaticMethodID(intClass, "valueOf", "(I)Ljava/lang/Integer;");
-  jobject jrate = env->CallStaticObjectMethod(intClass, intValueOf, sampleRate);
-  env->SetObjectArrayElement(result, 1, jrate);
-  env->DeleteLocalRef(intClass);
-  if (jrate) env->DeleteLocalRef(jrate);
-
-  env->DeleteLocalRef(objClass);
-  return result;
-}
-
-}  // namespace
-
-extern "C" {
-
-// Create a Zipvoice TTS instance via C-API. Returns the pointer as a jlong (0 on failure).
-JNIEXPORT jlong JNICALL
-Java_com_sherpaonnx_ZipvoiceTtsWrapper_nativeCreate(
-    JNIEnv* env, jobject /* this */,
-    jstring j_tokens, jstring j_encoder, jstring j_decoder, jstring j_vocoder,
-    jstring j_data_dir, jstring j_lexicon,
-    jfloat feat_scale, jfloat t_shift, jfloat target_rms, jfloat guidance_scale,
-    jint num_threads, jboolean debug,
-    jstring j_rule_fsts, jstring j_rule_fars, jint max_num_sentences, jfloat silence_scale,
-    jstring j_provider) {
-  JStringGuard tokens(env, j_tokens);
-  JStringGuard encoder(env, j_encoder);
-  JStringGuard decoder(env, j_decoder);
-  JStringGuard vocoder(env, j_vocoder);
-  JStringGuard dataDir(env, j_data_dir);
-  JStringGuard lexicon(env, j_lexicon);
-  JStringGuard ruleFsts(env, j_rule_fsts);
-  JStringGuard ruleFars(env, j_rule_fars);
-  JStringGuard provider(env, j_provider);
-
-  LOGI("nativeCreate: tokens=%s, encoder=%s, decoder=%s, vocoder=%s, dataDir=%s, lexicon=%s",
-       tokens.get(), encoder.get(), decoder.get(), vocoder.get(), dataDir.get(), lexicon.get());
-  LOGI("nativeCreate: featScale=%.3f, tShift=%.3f, targetRms=%.3f, guidanceScale=%.3f, threads=%d, debug=%d",
-       feat_scale, t_shift, target_rms, guidance_scale, num_threads, debug);
-  LOGI("nativeCreate: ruleFsts=%s, ruleFars=%s, maxNumSentences=%d, silenceScale=%.3f, provider=%s",
-       ruleFsts.get(), ruleFars.get(), max_num_sentences, silence_scale, provider.get());
-
-  SherpaOnnxOfflineTtsConfig config;
-  memset(&config, 0, sizeof(config));
-
-  config.model.zipvoice.tokens = tokens.get();
-  config.model.zipvoice.encoder = encoder.get();
-  config.model.zipvoice.decoder = decoder.get();
-  config.model.zipvoice.vocoder = vocoder.get();
-  config.model.zipvoice.data_dir = dataDir.get();
-  config.model.zipvoice.lexicon = lexicon.get();
-  config.model.zipvoice.feat_scale = feat_scale;
-  config.model.zipvoice.t_shift = t_shift;
-  config.model.zipvoice.target_rms = target_rms;
-  config.model.zipvoice.guidance_scale = guidance_scale;
-
-  config.model.num_threads = num_threads;
-  config.model.debug = debug ? 1 : 0;
-  config.model.provider = (provider.get() && *provider.get()) ? provider.get() : "cpu";
-
-  config.rule_fsts = ruleFsts.get();
-  config.rule_fars = ruleFars.get();
-  config.max_num_sentences = max_num_sentences;
-  config.silence_scale = silence_scale;
-
-  const SherpaOnnxOfflineTts* tts = SherpaOnnxCreateOfflineTts(&config);
-  if (!tts) {
-    LOGE("nativeCreate: SherpaOnnxCreateOfflineTts returned null");
-    return 0;
-  }
-
-  LOGI("nativeCreate: success, sampleRate=%d, numSpeakers=%d",
-       SherpaOnnxOfflineTtsSampleRate(tts), SherpaOnnxOfflineTtsNumSpeakers(tts));
-
-  return reinterpret_cast<jlong>(tts);
-}
-
-// Destroy a Zipvoice TTS instance.
-JNIEXPORT void JNICALL
-Java_com_sherpaonnx_ZipvoiceTtsWrapper_nativeDestroy(
-    JNIEnv* /* env */, jobject /* this */, jlong ptr) {
-  auto* tts = reinterpret_cast<const SherpaOnnxOfflineTts*>(ptr);
-  if (tts) {
-    SherpaOnnxDestroyOfflineTts(tts);
-    LOGI("nativeDestroy: released");
-  }
-}
-
-// Get the sample rate of the Zipvoice TTS model.
-JNIEXPORT jint JNICALL
-Java_com_sherpaonnx_ZipvoiceTtsWrapper_nativeGetSampleRate(
-    JNIEnv* /* env */, jobject /* this */, jlong ptr) {
-  auto* tts = reinterpret_cast<const SherpaOnnxOfflineTts*>(ptr);
-  return tts ? SherpaOnnxOfflineTtsSampleRate(tts) : 0;
-}
-
-// Get the number of speakers of the Zipvoice TTS model.
-JNIEXPORT jint JNICALL
-Java_com_sherpaonnx_ZipvoiceTtsWrapper_nativeGetNumSpeakers(
-    JNIEnv* /* env */, jobject /* this */, jlong ptr) {
-  auto* tts = reinterpret_cast<const SherpaOnnxOfflineTts*>(ptr);
-  return tts ? SherpaOnnxOfflineTtsNumSpeakers(tts) : 0;
-}
-
-// Generate audio (non-zero-shot). Returns Object[] { float[], Integer }.
-JNIEXPORT jobjectArray JNICALL
-Java_com_sherpaonnx_ZipvoiceTtsWrapper_nativeGenerate(
-    JNIEnv* env, jobject /* this */,
-    jlong ptr, jstring j_text, jint sid, jfloat speed) {
-  auto* tts = reinterpret_cast<const SherpaOnnxOfflineTts*>(ptr);
-  if (!tts) {
-    LOGE("nativeGenerate: tts pointer is null");
-    return nullptr;
-  }
-
-  JStringGuard text(env, j_text);
-  LOGI("nativeGenerate: text=%s, sid=%d, speed=%.2f", text.get(), sid, speed);
-
-  const SherpaOnnxGeneratedAudio* audio =
-      SherpaOnnxOfflineTtsGenerate(tts, text.get(), sid, speed);
-  if (!audio) {
-    LOGE("nativeGenerate: SherpaOnnxOfflineTtsGenerate returned null");
-    return nullptr;
-  }
-
-  LOGI("nativeGenerate: got %d samples at %d Hz", audio->n, audio->sample_rate);
-  jobjectArray result = buildAudioResult(env, audio->samples, audio->n, audio->sample_rate);
-
-  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
-  return result;
-}
-
-// Generate audio with callback for streaming. Returns Object[] { float[], Integer } for the
-// final concatenated audio. The callback is invoked per chunk.
-JNIEXPORT jobjectArray JNICALL
-Java_com_sherpaonnx_ZipvoiceTtsWrapper_nativeGenerateWithCallback(
-    JNIEnv* env, jobject thiz,
-    jlong ptr, jstring j_text, jint sid, jfloat speed) {
-  auto* tts = reinterpret_cast<const SherpaOnnxOfflineTts*>(ptr);
-  if (!tts) {
-    LOGE("nativeGenerateWithCallback: tts pointer is null");
-    return nullptr;
-  }
-
-  JStringGuard text(env, j_text);
-
-  // We use the progress callback variant to get chunks.
-  // The JNI environment and `thiz` are stored in a struct passed through void* arg.
-  struct CallbackCtx {
-    JNIEnv* env;
-    jobject thiz;
-    jmethodID onChunkId;
-    bool cancelled;
-  };
-
-  jclass cls = env->GetObjectClass(thiz);
-  jmethodID onChunkId = env->GetMethodID(cls, "onNativeChunk", "([FI)Z");
-  env->DeleteLocalRef(cls);
-  if (!onChunkId) {
-    LOGE("nativeGenerateWithCallback: onNativeChunk method not found");
-    return nullptr;
-  }
-
-  CallbackCtx ctx{env, thiz, onChunkId, false};
-
-  auto callback = [](const float* samples, int32_t n, float /* progress */, void* arg) -> int32_t {
-    auto* c = static_cast<CallbackCtx*>(arg);
-    if (c->cancelled) return 0;
-
-    jfloatArray chunk = c->env->NewFloatArray(n);
-    if (chunk && n > 0) {
-      c->env->SetFloatArrayRegion(chunk, 0, n, samples);
-    }
-
-    // Call Java: boolean onNativeChunk(float[] samples, int n)
-    jboolean cont = c->env->CallBooleanMethod(c->thiz, c->onChunkId, chunk, n);
-    if (chunk) c->env->DeleteLocalRef(chunk);
-
-    if (!cont) {
-      c->cancelled = true;
-      return 0;
-    }
-    return 1;
-  };
-
-  const SherpaOnnxGeneratedAudio* audio =
-      SherpaOnnxOfflineTtsGenerateWithProgressCallbackWithArg(
-          tts, text.get(), sid, speed, callback, &ctx);
-
-  if (!audio) {
-    LOGE("nativeGenerateWithCallback: generate returned null");
-    return nullptr;
-  }
-
-  jobjectArray result = buildAudioResult(env, audio->samples, audio->n, audio->sample_rate);
-  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
-  return result;
-}
-
-// Zero-shot voice cloning with Zipvoice. Returns Object[] { float[], Integer }.
-JNIEXPORT jobjectArray JNICALL
-Java_com_sherpaonnx_ZipvoiceTtsWrapper_nativeGenerateWithZipvoice(
-    JNIEnv* env, jobject /* this */,
-    jlong ptr, jstring j_text, jstring j_prompt_text,
-    jfloatArray j_prompt_samples, jint prompt_sr,
-    jfloat speed, jint num_steps) {
-  auto* tts = reinterpret_cast<const SherpaOnnxOfflineTts*>(ptr);
-  if (!tts) {
-    LOGE("nativeGenerateWithZipvoice: tts pointer is null");
-    return nullptr;
-  }
-
-  JStringGuard text(env, j_text);
-  JStringGuard promptText(env, j_prompt_text);
-
-  jfloat* promptSamples = nullptr;
-  jint nPrompt = 0;
-  if (j_prompt_samples) {
-    nPrompt = env->GetArrayLength(j_prompt_samples);
-    promptSamples = env->GetFloatArrayElements(j_prompt_samples, nullptr);
-  }
-
-  LOGI("nativeGenerateWithZipvoice: text=%s, promptLen=%d, promptSr=%d, speed=%.2f, steps=%d",
-       text.get(), nPrompt, prompt_sr, speed, num_steps);
-
-  const SherpaOnnxGeneratedAudio* audio =
-      SherpaOnnxOfflineTtsGenerateWithZipvoice(
-          tts, text.get(), promptText.get(),
-          promptSamples, nPrompt, prompt_sr,
-          speed, num_steps);
-
-  if (promptSamples) {
-    env->ReleaseFloatArrayElements(j_prompt_samples, promptSamples, JNI_ABORT);
-  }
-
-  if (!audio) {
-    LOGE("nativeGenerateWithZipvoice: returned null");
-    return nullptr;
-  }
-
-  LOGI("nativeGenerateWithZipvoice: got %d samples at %d Hz", audio->n, audio->sample_rate);
-  jobjectArray result = buildAudioResult(env, audio->samples, audio->n, audio->sample_rate);
-
-  SherpaOnnxDestroyOfflineTtsGeneratedAudio(audio);
-  return result;
-}
-
-}  // extern "C"