react-native-sherpa-onnx 0.4.0 → 0.4.1

package/android/src/main/java/com/sherpaonnx/SherpaOnnxTtsHelper.kt

@@ -111,6 +111,46 @@ internal class SherpaOnnxTtsHelper(
     }
   }
 
+  /**
+   * libsherpa-onnx-jni looks up `invoke([F)Ljava/lang/Integer` (see sherpa-onnx `offline-tts.cc` CallCallback).
+   * Kotlin `Function1<*, Int>` compiles to `invoke([F)I`, so GetMethodID fails and JNI aborts.
+   * Using [java.lang.Integer] as the type parameter yields the boxed JVM signature the JNI expects.
+   * The cast is only for the Kotlin API (`generateWithCallback` still declares `Function1<FloatArray, Int>`).
+   */
+  /** Box for JNI: must be real [java.lang.Integer], not Kotlin [Int] (primitive `invoke([F)I` breaks sherpa JNI). */
+  @Suppress("DEPRECATION")
+  private fun boxForTtsJni(n: Int): java.lang.Integer = java.lang.Integer(n)
+
+  @Suppress("UNCHECKED_CAST")
+  private fun ttsChunkCallbackForJni(
+    sentenceChunkSizes: MutableList<Int>
+  ): kotlin.Function1<FloatArray, Int> {
+    val boxed =
+      object : kotlin.jvm.functions.Function1<FloatArray, java.lang.Integer> {
+        override fun invoke(chunk: FloatArray): java.lang.Integer {
+          sentenceChunkSizes.add(chunk.size)
+          return boxForTtsJni(chunk.size)
+        }
+      }
+    return boxed as kotlin.Function1<FloatArray, Int>
+  }
+
+  @Suppress("UNCHECKED_CAST")
+  private fun ttsStreamChunkCallbackForJni(
+    cancelled: AtomicBoolean,
+    onChunk: (FloatArray) -> Unit
+  ): kotlin.Function1<FloatArray, Int> {
+    val boxed =
+      object : kotlin.jvm.functions.Function1<FloatArray, java.lang.Integer> {
+        override fun invoke(chunk: FloatArray): java.lang.Integer {
+          if (cancelled.get()) return boxForTtsJni(0)
+          onChunk(chunk)
+          return boxForTtsJni(chunk.size)
+        }
+      }
+    return boxed as kotlin.Function1<FloatArray, Int>
+  }
+
   /** Single-thread executor for TTS init so the RN bridge thread is not blocked (avoids Inspector/dev WebSocket races in debug builds). */
   private val ttsInitExecutor = Executors.newSingleThreadExecutor()
 
@@ -453,6 +493,7 @@ internal class SherpaOnnxTtsHelper(
       }
       val sid = getSid(options)
       val speed = getSpeed(options)
+      val sentenceChunkSizes = mutableListOf<Int>()
       val audio = when {
         hasReferenceAudio(options) && (inst.isZipvoice || inst.isPocket) -> {
           if (inst.isZipvoice) {
@@ -467,7 +508,11 @@ internal class SherpaOnnxTtsHelper(
             }
           }
           val config = parseGenerationConfig(options) ?: GenerationConfig(speed = speed, sid = sid)
-          inst.tts!!.generateWithConfig(text, config)
+          inst.tts!!.generateWithConfigAndCallback(
+            text,
+            config,
+            ttsChunkCallbackForJni(sentenceChunkSizes)
+          )
         }
         hasReferenceAudio(options) -> {
           Log.e("SherpaOnnxTts", "TTS_GENERATE_ERROR: Reference audio is not supported for this TTS model type")
@@ -485,12 +530,15 @@ internal class SherpaOnnxTtsHelper(
           )
           return
         }
-        else -> dispatchGenerate(inst, text, sid, speed)
-          ?: run {
+        else -> {
+          val tts = inst.tts
+          if (tts == null) {
             Log.e("SherpaOnnxTts", "TTS_GENERATE_ERROR: TTS not initialized")
             promise.reject("TTS_GENERATE_ERROR", "TTS not initialized")
             return
           }
+          tts.generateWithCallback(text, sid, speed, ttsChunkCallbackForJni(sentenceChunkSizes))
+        }
       }
       val map = Arguments.createMap()
       val samplesArray = Arguments.createArray()
@@ -564,18 +612,23 @@ internal class SherpaOnnxTtsHelper(
         when {
           hasReferenceAudio(options) && inst.isPocket -> {
             val config = parseGenerationConfig(options) ?: GenerationConfig(speed = speed, sid = sid)
-            inst.tts!!.generateWithConfigAndCallback(text, config) { chunk ->
-              if (inst.ttsStreamCancelled.get()) return@generateWithConfigAndCallback 0
-              emitChunk(instanceId, requestId, chunk, sampleRate, 0f, false)
-              chunk.size
-            }
+            inst.tts!!.generateWithConfigAndCallback(
+              text,
+              config,
+              ttsStreamChunkCallbackForJni(inst.ttsStreamCancelled) { chunk ->
+                emitChunk(instanceId, requestId, chunk, sampleRate, 0f, false)
+              }
+            )
           }
           else -> {
-            inst.tts!!.generateWithCallback(text, sid, speed) { chunk ->
-              if (inst.ttsStreamCancelled.get()) return@generateWithCallback 0
-              emitChunk(instanceId, requestId, chunk, sampleRate, 0f, false)
-              chunk.size
-            }
+            inst.tts!!.generateWithCallback(
+              text,
+              sid,
+              speed,
+              ttsStreamChunkCallbackForJni(inst.ttsStreamCancelled) { chunk ->
+                emitChunk(instanceId, requestId, chunk, sampleRate, 0f, false)
+              }
+            )
           }
         }
         if (!inst.ttsStreamCancelled.get()) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "react-native-sherpa-onnx",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "description": "Offline Speech-to-text, text-to-speech, speaker diarization, speech enhancement, source separation, and VAD with sherpa-onnx for React NativeSpeech-to-Text with sherpa-onnx for React Native",
   "main": "./lib/module/index.js",
   "types": "./lib/typescript/src/index.d.ts",