Add JavaScript API (WebAssembly) for Moonshine v2 models (#3241)

This pull request expands the sherpa-onnx library's capabilities by adding comprehensive support for Moonshine v2 models within its JavaScript/WebAssembly API. This includes necessary modifications to model configurations, the introduction of a new example for practical usage, and updates to the continuous integration pipeline to ensure stability and correctness.

.github/scripts/test-nodejs-npm.sh

@@ -9,6 +9,16 @@ git status
 ls -lh
 ls -lh node_modules
 
+echo "---test moonshine v2---"
+
+curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27.tar.bz2
+tar xvf sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27.tar.bz2
+rm sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27.tar.bz2
+
+node ./test-offline-moonshine-v2.js
+
+rm -rf sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27
+
 echo "---test FireRedASR CTC---"
 curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-fire-red-asr2-ctc-zh_en-int8-2026-02-25.tar.bz2
 tar xvf sherpa-onnx-fire-red-asr2-ctc-zh_en-int8-2026-02-25.tar.bz2

nodejs-examples/README.md

@@ -422,6 +422,22 @@ rm sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16.tar.bz2
 node ./test-offline-fire-red-asr.js
 ```
 
+## ./test-offline-moonshine-v2.js
+
+[./test-offline-moonshine-v2.js](./test-offline-moonshine-v2.js) demonstrates
+how to decode a file with a Moonshine v2 model. In the code we use
+[sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27](https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27.tar.bz2).
+
+You can use the following command to run it:
+
+```bash
+wget https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27.tar.bz2
+tar xvf sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27.tar.bz2
+rm sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27.tar.bz2
+
+node ./test-offline-moonshine-v2.js
+```
+
 ## ./test-offline-moonshine.js
 
 [./test-offline-moonshine.js](./test-offline-moonshine.js) demonstrates

nodejs-examples/test-offline-fire-red-asr.js

@@ -21,8 +21,8 @@ function createOfflineRecognizer() {
   return sherpa_onnx.createOfflineRecognizer(config);
 }
 
-recognizer = createOfflineRecognizer();
-stream = recognizer.createStream();
+const recognizer = createOfflineRecognizer();
+const stream = recognizer.createStream();
 
 const waveFilename =
     './sherpa-onnx-fire-red-asr-large-zh_en-2025-02-16/test_wavs/0.wav';

nodejs-examples/test-offline-moonshine-v2.js

@@ -0,0 +1,36 @@
+// Copyright (c)  2023-2026  Xiaomi Corporation (authors: Fangjun Kuang)
+//
+const sherpa_onnx = require('sherpa-onnx');
+
+function createOfflineRecognizer() {
+  let modelConfig = {
+    moonshine: {
+      encoder:
+          './sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27/encoder_model.ort',
+      mergedDecoder:
+          './sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27/decoder_model_merged.ort',
+    },
+    tokens: './sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27/tokens.txt',
+  };
+
+  let config = {
+    modelConfig: modelConfig,
+  };
+
+  return sherpa_onnx.createOfflineRecognizer(config);
+}
+
+const recognizer = createOfflineRecognizer();
+const stream = recognizer.createStream();
+
+const waveFilename =
+    './sherpa-onnx-moonshine-tiny-en-quantized-2026-02-27/test_wavs/0.wav';
+const wave = sherpa_onnx.readWave(waveFilename);
+stream.acceptWaveform(wave.sampleRate, wave.samples);
+
+recognizer.decode(stream);
+const text = recognizer.getResult(stream).text;
+console.log(text);
+
+stream.free();
+recognizer.free();

nodejs-examples/test-offline-moonshine.js

@@ -22,8 +22,8 @@ function createOfflineRecognizer() {
   return sherpa_onnx.createOfflineRecognizer(config);
 }
 
-recognizer = createOfflineRecognizer();
-stream = recognizer.createStream();
+const recognizer = createOfflineRecognizer();
+const stream = recognizer.createStream();
 
 const waveFilename = './sherpa-onnx-moonshine-tiny-en-int8/test_wavs/0.wav';
 const wave = sherpa_onnx.readWave(waveFilename);

nodejs-examples/test-offline-speech-enhancement-gtcrn.js

@@ -18,7 +18,7 @@ function createOfflineSpeechDenoiser() {
   return sherpa_onnx.createOfflineSpeechDenoiser(config);
 }
 
-speech_denoiser = createOfflineSpeechDenoiser();
+const speech_denoiser = createOfflineSpeechDenoiser();
 
 const waveFilename = './inp_16k.wav';
 const wave = sherpa_onnx.readWave(waveFilename);

nodejs-examples/test-offline-whisper.js

@@ -24,8 +24,8 @@ function createOfflineRecognizer() {
   return sherpa_onnx.createOfflineRecognizer(config);
 }
 
-recognizer = createOfflineRecognizer();
-stream = recognizer.createStream();
+const recognizer = createOfflineRecognizer();
+const stream = recognizer.createStream();
 
 const waveFilename = './sherpa-onnx-whisper-tiny.en/test_wavs/0.wav';
 const wave = sherpa_onnx.readWave(waveFilename);

nodejs-examples/test-online-paraformer-microphone-mic.js

@@ -1,11 +1,15 @@
 // Copyright (c) 2023 Xiaomi Corporation (authors: Fangjun Kuang)
-const mic = require('mic'); // It uses `mic` for better compatibility, do check its [npm](https://www.npmjs.com/package/mic) before running it.
+const mic = require(
+    'mic');  // It uses `mic` for better compatibility, do check its
+             // [npm](https://www.npmjs.com/package/mic) before running it.
 const sherpa_onnx = require('sherpa-onnx');
 
 function createOnlineRecognizer() {
   let onlineParaformerModelConfig = {
-    encoder: './sherpa-onnx-streaming-paraformer-bilingual-zh-en/encoder.int8.onnx',
-    decoder: './sherpa-onnx-streaming-paraformer-bilingual-zh-en/decoder.int8.onnx',
+    encoder:
+        './sherpa-onnx-streaming-paraformer-bilingual-zh-en/encoder.int8.onnx',
+    decoder:
+        './sherpa-onnx-streaming-paraformer-bilingual-zh-en/decoder.int8.onnx',
   };
 
   let onlineModelConfig = {
@@ -62,32 +66,32 @@ class SpeechSession {
   }
 
   shouldStartNewSession() {
-    return Date.now() - this.lastUpdateTime > 10000; // 10 seconds of silence
+    return Date.now() - this.lastUpdateTime > 10000;  // 10 seconds of silence
   }
 }
 
 function formatOutput() {
-    clearConsole();
-    console.log('\n=== Automated Speech Recognition ===');
-    console.log(`Current Session #${sessionCount}`);
-    console.log('Time:', new Date().toLocaleTimeString());
+  clearConsole();
+  console.log('\n=== Automated Speech Recognition ===');
+  console.log(`Current Session #${sessionCount}`);
+  console.log('Time:', new Date().toLocaleTimeString());
+  console.log('------------------------');
+
+  // 显示历史句子
+  if (currentSession.sentences.length > 0) {
+    console.log('Recognized Sentences:');
+    currentSession.sentences.forEach((sentence, index) => {
+      console.log(`[${sentence.timestamp}] ${index + 1}. ${sentence.text}`);
+    });
     console.log('------------------------');
-    
-    // 显示历史句子
-    if (currentSession.sentences.length > 0) {
-      console.log('Recognized Sentences:');
-      currentSession.sentences.forEach((sentence, index) => {
-        console.log(`[${sentence.timestamp}] ${index + 1}. ${sentence.text}`);
-      });
-      console.log('------------------------');
-    }
-    
-    // 显示当前正在识别的内容
-    if (currentSession.currentText) {
-      console.log('Recognizing:', currentSession.currentText);
-    }
   }
-  
+
+  // 显示当前正在识别的内容
+  if (currentSession.currentText) {
+    console.log('Recognizing:', currentSession.currentText);
+  }
+}
+
 
 const recognizer = createOnlineRecognizer();
 const stream = recognizer.createStream();
@@ -113,7 +117,7 @@ function exitHandler(options, exitCode) {
 const micInstance = mic({
   rate: recognizer.config.featConfig.sampleRate,
   channels: 1,
-  debug: false, // 关闭调试输出
+  debug: false,  // 关闭调试输出
   device: 'default',
   bitwidth: 16,
   encoding: 'signed-integer',
@@ -129,12 +133,12 @@ function startMic() {
       console.log('Mic phone started.');
       resolve();
     });
-    
+
     micInputStream.once('error', (err) => {
       console.error('Mic phone start error:', err);
       reject(err);
     });
-    
+
     micInstance.start();
   });
 }
@@ -142,7 +146,7 @@ function startMic() {
 micInputStream.on('data', buffer => {
   const int16Array = new Int16Array(buffer.buffer);
   const samples = new Float32Array(int16Array.length);
-  
+
   for (let i = 0; i < int16Array.length; i++) {
     samples[i] = int16Array[i] / 32768.0;
   }
@@ -163,7 +167,7 @@ micInputStream.on('data', buffer => {
       sessionCount++;
       currentSession = new SpeechSession();
     }
-    
+
     currentSession.addOrUpdateText(text);
     formatOutput();
   }
@@ -203,4 +207,4 @@ async function main() {
   }
 }
 
-main();
+main();

wasm/asr/sherpa-onnx-asr.js

@@ -1050,12 +1050,14 @@ function initSherpaOnnxOfflineMoonshineModelConfig(config, Module) {
       Module.lengthBytesUTF8(config.uncachedDecoder || '') + 1;
   const cachedDecoderLen =
       Module.lengthBytesUTF8(config.cachedDecoder || '') + 1;
+  const mergedDecoderLen =
+      Module.lengthBytesUTF8(config.mergedDecoder || '') + 1;
 
-  const n =
-      preprocessorLen + encoderLen + uncachedDecoderLen + cachedDecoderLen;
+  const n = preprocessorLen + encoderLen + uncachedDecoderLen +
+      cachedDecoderLen + mergedDecoderLen;
   const buffer = Module._malloc(n);
 
-  const len = 4 * 4;  // 4 pointers
+  const len = 5 * 4;  // 5 pointers
   const ptr = Module._malloc(len);
 
   let offset = 0;
@@ -1074,6 +1076,10 @@ function initSherpaOnnxOfflineMoonshineModelConfig(config, Module) {
       config.cachedDecoder || '', buffer + offset, cachedDecoderLen);
   offset += cachedDecoderLen;
 
+  Module.stringToUTF8(
+      config.mergedDecoder || '', buffer + offset, mergedDecoderLen);
+  offset += mergedDecoderLen;
+
   offset = 0;
   Module.setValue(ptr, buffer + offset, 'i8*');
   offset += preprocessorLen;
@@ -1087,6 +1093,9 @@ function initSherpaOnnxOfflineMoonshineModelConfig(config, Module) {
   Module.setValue(ptr + 12, buffer + offset, 'i8*');
   offset += cachedDecoderLen;
 
+  Module.setValue(ptr + 16, buffer + offset, 'i8*');
+  offset += mergedDecoderLen;
+
   return {
     buffer: buffer,
     ptr: ptr,
@@ -1288,6 +1297,7 @@ function initSherpaOnnxOfflineModelConfig(config, Module) {
       encoder: '',
       uncachedDecoder: '',
       cachedDecoder: '',
+      mergedDecoder: '',
     };
   }
 

wasm/nodejs/sherpa-onnx-wasm-nodejs.cc

@@ -23,7 +23,7 @@ static_assert(sizeof(SherpaOnnxOfflineDolphinModelConfig) == 4, "");
 static_assert(sizeof(SherpaOnnxOfflineNemoEncDecCtcModelConfig) == 4, "");
 static_assert(sizeof(SherpaOnnxOfflineWhisperModelConfig) == 7 * 4, "");
 static_assert(sizeof(SherpaOnnxOfflineFireRedAsrModelConfig) == 2 * 4, "");
-static_assert(sizeof(SherpaOnnxOfflineMoonshineModelConfig) == 4 * 4, "");
+static_assert(sizeof(SherpaOnnxOfflineMoonshineModelConfig) == 5 * 4, "");
 static_assert(sizeof(SherpaOnnxOfflineTdnnModelConfig) == 4, "");
 static_assert(sizeof(SherpaOnnxOfflineSenseVoiceModelConfig) == 3 * 4, "");
 static_assert(sizeof(SherpaOnnxOfflineCanaryModelConfig) == 5 * 4, "");
@@ -130,6 +130,7 @@ void PrintOfflineRecognizerConfig(SherpaOnnxOfflineRecognizerConfig *config) {
   fprintf(stdout, "encoder: %s\n", moonshine->encoder);
   fprintf(stdout, "uncached_decoder: %s\n", moonshine->uncached_decoder);
   fprintf(stdout, "cached_decoder: %s\n", moonshine->cached_decoder);
+  fprintf(stdout, "merged_decoder: %s\n", moonshine->merged_decoder);
 
   fprintf(stdout, "----------offline FireRedAsr model config----------\n");
   fprintf(stdout, "encoder: %s\n", fire_red_asr->encoder);