Add Go API for Moonshine models (#1479)

.github/workflows/test-go.yaml

@@ -134,6 +134,53 @@ jobs:
           name: ${{ matrix.os }}-libs
           path: to-upload/
 
+      - name: Test non-streaming decoding files
+        shell: bash
+        run: |
+          cd scripts/go/_internal/non-streaming-decode-files/
+          ls -lh
+          go mod tidy
+          cat go.mod
+          go build
+          ls -lh
+
+          echo "Test Moonshine"
+          ./run-moonshine.sh
+          rm -rf sherpa-onnx-*
+
+          echo "Test SenseVoice ctc"
+          ./run-sense-voice-small.sh
+          rm -rf sherpa-onnx-sense-*
+
+          echo "Test telespeech ctc"
+          ./run-telespeech-ctc.sh
+          rm -rf sherpa-onnx-telespeech-ctc-*
+
+          echo "Test transducer"
+          ./run-transducer.sh
+          rm -rf sherpa-onnx-zipformer-en-2023-06-26
+
+          echo "Test transducer"
+          ./run-transducer.sh
+          rm -rf sherpa-onnx-zipformer-en-2023-06-26
+
+          echo "Test paraformer"
+          ./run-paraformer.sh
+          ./run-paraformer-itn.sh
+          rm -rf sherpa-onnx-paraformer-zh-2023-09-14
+
+          echo "Test NeMo CTC"
+          ./run-nemo-ctc.sh
+          rm -rf sherpa-onnx-nemo-ctc-en-conformer-medium
+
+          echo "Test Whisper tiny.en"
+          ./run-whisper.sh
+          rm -rf sherpa-onnx-whisper-tiny.en
+
+          echo "Test Tdnn yesno"
+          ./run-tdnn-yesno.sh
+          rm -rf sherpa-onnx-tdnn-yesno
+
       - name: Test adding punctuation
         shell: bash
         run: |
@@ -193,49 +240,6 @@ jobs:
           name: tts-waves-${{ matrix.os }}
           path: tts-waves
 
-      - name: Test non-streaming decoding files
-        shell: bash
-        run: |
-          cd scripts/go/_internal/non-streaming-decode-files/
-          ls -lh
-          go mod tidy
-          cat go.mod
-          go build
-          ls -lh
-
-          echo "Test SenseVoice ctc"
-          ./run-sense-voice-small.sh
-          rm -rf sherpa-onnx-sense-*
-
-          echo "Test telespeech ctc"
-          ./run-telespeech-ctc.sh
-          rm -rf sherpa-onnx-telespeech-ctc-*
-
-          echo "Test transducer"
-          ./run-transducer.sh
-          rm -rf sherpa-onnx-zipformer-en-2023-06-26
-
-          echo "Test transducer"
-          ./run-transducer.sh
-          rm -rf sherpa-onnx-zipformer-en-2023-06-26
-
-          echo "Test paraformer"
-          ./run-paraformer.sh
-          ./run-paraformer-itn.sh
-          rm -rf sherpa-onnx-paraformer-zh-2023-09-14
-
-          echo "Test NeMo CTC"
-          ./run-nemo-ctc.sh
-          rm -rf sherpa-onnx-nemo-ctc-en-conformer-medium
-
-          echo "Test Whisper tiny.en"
-          ./run-whisper.sh
-          rm -rf sherpa-onnx-whisper-tiny.en
-
-          echo "Test Tdnn yesno"
-          ./run-tdnn-yesno.sh
-          rm -rf sherpa-onnx-tdnn-yesno
-
       - name: Test streaming decoding files
         shell: bash
         run: |

go-api-examples/README.md

@@ -6,28 +6,41 @@ Please refer to the documentation
 https://k2-fsa.github.io/sherpa/onnx/go-api/index.html
 for details.
 
+- [./add-punctuation](./add-punctuation) It shows how to use
+  a punctuation model to add punctuations to text
+
 - [./non-streaming-decode-files](./non-streaming-decode-files) It shows how to use
   a non-streaming ASR model to decode files
 
+- [./non-streaming-speaker-diarization](./non-streaming-speaker-diarization) It shows how to use
+  a speaker segmentation model and a speaker embedding model for speaker diarization.
+
 - [./non-streaming-tts](./non-streaming-tts) It shows how to use a non-streaming TTS
   model to convert text to speech
 
 - [./real-time-speech-recognition-from-microphone](./real-time-speech-recognition-from-microphone)
   It shows how to use a streaming ASR model to recognize speech from a microphone in real-time
 
+- [./speaker-identification](./speaker-identification) It shows how to use a speaker
+  embedding model for speaker identification.
+
+- [./streaming-decode-files](./streaming-decode-files) It shows how to use a streaming
+  model for streaming speech recognition
+
+- [./streaming-hlg-decoding](./streaming-hlg-decoding) It shows how to use a streaming
+  model for streaming speech recognition with HLG decoding
+
 - [./vad](./vad) It shows how to use silero VAD with Golang.
 
-- [./vad-asr-whisper](./vad-asr-whisper) It shows how to use silero VAD + Whisper
+- [./vad-asr-paraformer](./vad-asr-paraformer) It shows how to use silero VAD + Paraformer
   for speech recognition.
 
-- [./vad-asr-paraformer](./vad-asr-paraformer) It shows how to use silero VAD + Paraformer
+- [./vad-asr-whisper](./vad-asr-whisper) It shows how to use silero VAD + Whisper
+
+- [./vad-speaker-identification](./vad-speaker-identification) It shows how to use Go API for VAD + speaker identification.
   for speech recognition.
 
 - [./vad-spoken-language-identification](./vad-spoken-language-identification) It shows how to use silero VAD + Whisper
   for spoken language identification.
 
-- [./speaker-identification](./speaker-identification) It shows how to use Go API for speaker identification.
-
-- [./vad-speaker-identification](./vad-speaker-identification) It shows how to use Go API for VAD + speaker identification.
-
 [sherpa-onnx]: https://github.com/k2-fsa/sherpa-onnx

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

@@ -34,6 +34,11 @@ func main() {
 	flag.StringVar(&config.ModelConfig.Whisper.Task, "whisper-task", "transcribe", "transcribe or translate")
 	flag.IntVar(&config.ModelConfig.Whisper.TailPaddings, "whisper-tail-paddings", -1, "tail paddings for whisper")
 
+	flag.StringVar(&config.ModelConfig.Moonshine.Preprocessor, "moonshine-preprocessor", "", "Path to the moonshine preprocessor model")
+	flag.StringVar(&config.ModelConfig.Moonshine.Encoder, "moonshine-encoder", "", "Path to the moonshine encoder model")
+	flag.StringVar(&config.ModelConfig.Moonshine.UncachedDecoder, "moonshine-uncached-decoder", "", "Path to the moonshine uncached decoder model")
+	flag.StringVar(&config.ModelConfig.Moonshine.CachedDecoder, "moonshine-cached-decoder", "", "Path to the moonshine cached decoder model")
+
 	flag.StringVar(&config.ModelConfig.Tdnn.Model, "tdnn-model", "", "Path to the tdnn model")
 
 	flag.StringVar(&config.ModelConfig.SenseVoice.Model, "sense-voice-model", "", "Path to the SenseVoice model")
@@ -85,12 +90,8 @@ func main() {
 	log.Println("Emotion: " + result.Emotion)
 	log.Println("Lang: " + result.Lang)
 	log.Println("Event: " + result.Event)
-	for _, v := range result.Timestamps {
-		log.Printf("Timestamp: %+v\n", v)
-	}
-	for _, v := range result.Tokens {
-		log.Println("Token: " + v)
-	}
+	log.Printf("Timestamp: %v\n", result.Timestamps)
+	log.Printf("Tokens: %v\n", result.Tokens)
 	log.Printf("Wave duration: %v seconds", float32(len(samples))/float32(sampleRate))
 }
 

go-api-examples/non-streaming-decode-files/run-moonshine.sh

@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+
+set -ex
+
+if [ ! -f ./sherpa-onnx-moonshine-tiny-en-int8/tokens.txt ]; then
+  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-moonshine-tiny-en-int8.tar.bz2
+  tar xvf sherpa-onnx-moonshine-tiny-en-int8.tar.bz2
+  rm sherpa-onnx-moonshine-tiny-en-int8.tar.bz2
+fi
+
+go mod tidy
+go build
+
+./non-streaming-decode-files \
+  --moonshine-preprocessor=./sherpa-onnx-moonshine-tiny-en-int8/preprocess.onnx \
+  --moonshine-encoder=./sherpa-onnx-moonshine-tiny-en-int8/encode.int8.onnx \
+  --moonshine-uncached-decoder=./sherpa-onnx-moonshine-tiny-en-int8/uncached_decode.int8.onnx \
+  --moonshine-cached-decoder=./sherpa-onnx-moonshine-tiny-en-int8/cached_decode.int8.onnx \
+  --tokens=./sherpa-onnx-moonshine-tiny-en-int8/tokens.txt \
+  ./sherpa-onnx-moonshine-tiny-en-int8/test_wavs/0.wav
+

scripts/go/_internal/non-streaming-decode-files/run-moonshine.sh

@@ -0,0 +1 @@
+../../../../go-api-examples/non-streaming-decode-files/run-moonshine.sh

scripts/go/sherpa_onnx.go

@@ -382,6 +382,13 @@ type OfflineWhisperModelConfig struct {
 	TailPaddings int
 }
 
+type OfflineMoonshineModelConfig struct {
+	Preprocessor    string
+	Encoder         string
+	UncachedDecoder string
+	CachedDecoder   string
+}
+
 type OfflineTdnnModelConfig struct {
 	Model string
 }
@@ -405,6 +412,7 @@ type OfflineModelConfig struct {
 	Whisper    OfflineWhisperModelConfig
 	Tdnn       OfflineTdnnModelConfig
 	SenseVoice OfflineSenseVoiceModelConfig
+	Moonshine  OfflineMoonshineModelConfig
 	Tokens     string // Path to tokens.txt
 
 	// Number of threads to use for neural network computation
@@ -515,6 +523,18 @@ func NewOfflineRecognizer(config *OfflineRecognizerConfig) *OfflineRecognizer {
 
 	c.model_config.sense_voice.use_itn = C.int(config.ModelConfig.SenseVoice.UseInverseTextNormalization)
 
+	c.model_config.moonshine.preprocessor = C.CString(config.ModelConfig.Moonshine.Preprocessor)
+	defer C.free(unsafe.Pointer(c.model_config.moonshine.preprocessor))
+
+	c.model_config.moonshine.encoder = C.CString(config.ModelConfig.Moonshine.Encoder)
+	defer C.free(unsafe.Pointer(c.model_config.moonshine.encoder))
+
+	c.model_config.moonshine.uncached_decoder = C.CString(config.ModelConfig.Moonshine.UncachedDecoder)
+	defer C.free(unsafe.Pointer(c.model_config.moonshine.uncached_decoder))
+
+	c.model_config.moonshine.cached_decoder = C.CString(config.ModelConfig.Moonshine.CachedDecoder)
+	defer C.free(unsafe.Pointer(c.model_config.moonshine.cached_decoder))
+
 	c.model_config.tokens = C.CString(config.ModelConfig.Tokens)
 	defer C.free(unsafe.Pointer(c.model_config.tokens))