Sherpa Go语言实战

# Note: You need to run pip install sherpa-onnx to get the commandline tool: sherpa-onnx-cli


sherpa-onnx-cli text2token --help
Usage: sherpa-onnx-cli text2token [OPTIONS] INPUT OUTPUT

Options:

  --text TEXT         Path to the input texts. Each line in the texts contains the original phrase, it might also contain some extra items,
                      for example, the boosting score (startting with :), the triggering threshold
                      (startting with #, only used in keyword spotting task) and the original phrase (startting with @).
                      Note: extra items will be kept in the output.

                      example input 1 (tokens_type = ppinyin):
                          小爱同学 :2.0 #0.6 @小爱同学
                          你好问问 :3.5 @你好问问
                          小艺小艺 #0.6 @小艺小艺
                      example output 1:
                          x iǎo ài tóng x ué :2.0 #0.6 @小爱同学
                          n ǐ h ǎo w èn w èn :3.5 @你好问问
                          x iǎo y ì x iǎo y ì #0.6 @小艺小艺

                      example input 2 (tokens_type = bpe):
                          HELLO WORLD :1.5 #0.4
                          HI GOOGLE :2.0 #0.8
                          HEY SIRI #0.35
                      example output 2:
                          ▁HE LL O ▁WORLD :1.5 #0.4
                          ▁HI ▁GO O G LE :2.0 #0.8
                          ▁HE Y ▁S I RI #0.35

  --tokens TEXT       The path to tokens.txt.
  --tokens-type TEXT  The type of modeling units, should be cjkchar, bpe, cjkchar+bpe, fpinyin or ppinyin.
                      fpinyin means full pinyin, each cjkchar has a pinyin(with tone). ppinyin
                      means partial pinyin, it splits pinyin into initial and final,
  --bpe-model TEXT    The path to bpe.model. Only required when tokens-type is bpe or cjkchar+bpe.
  --help              Show this message and exit.

  	import "github.com/gordonklaus/portaudio" 

func (t *ASRServer) SetupEngine(ctx context.Context) {
	lastText := ""
    framesPerBuffer := 1024
	err := portaudio.Initialize()
	if err != nil {
		t.logger.Fatal().Err(err).Msg("cannot init portaudio")
	}
	defer portaudio.Terminate()

	defaultDevice, err := portaudio.DefaultInputDevice()
	if err != nil {
		t.logger.Fatal().Err(err).Msg("failed to get default input device")
	}

	portAudioParam := portaudio.StreamParameters{
		Input: portaudio.StreamDeviceParameters{
			Device:   defaultDevice,
			Channels: 1,
			Latency:  defaultDevice.DefaultLowInputLatency,
		},
		SampleRate:      16000,
		FramesPerBuffer: framesPerBuffer,
		Flags:           portaudio.ClipOff,
	}
	config := createNewOnlineRecognizerConfig()

	recognizer := sherpa.NewOnlineRecognizer(&config)
	defer sherpa.DeleteOnlineRecognizer(recognizer)

	stream := sherpa.NewOnlineStream(recognizer)
	defer sherpa.DeleteOnlineStream(stream)

	// 每次采样的时长
	samplesPerCall := int32(framesPerBuffer) 

	samples := make([]float32, samplesPerCall)
	s, err := portaudio.OpenStream(portAudioParam, samples)
	if err != nil {
		t.logger.Fatal().Err(err).Msg("failed to open stream")
	}
	defer s.Stop()
	defer s.Close()

	err = s.Start()
	if err != nil {
		t.logger.Fatal().Err(err).Msg("failed to start stream")
	}

	t.logger.Info().Msg("service is ready..")

	for {
		select {
		case <-ctx.Done():
			break
		default:
			if t.paused.Load() {
				continue
			}
			err = s.Read()
			if err != nil {
				t.logger.Error().Err(err).Msg("failed to read")
				continue
			}
			stream.AcceptWaveform(config.FeatConfig.SampleRate, samples)
			for recognizer.IsReady(stream) {
				recognizer.Decode(stream)
			}
			text := recognizer.GetResult(stream).Text
			if len(text) != 0 && lastText != text {
				lastText = strings.ToLower(text)
			}
			if recognizer.IsEndpoint(stream) {
				if len(text) != 0 {
					t.logger.Info().Str("text", lastText).Msg("recognized text")
					t.subscribers.Range(func(ch, _ any) bool {
						if channel, ok := ch.(chan string); ok {
							select {
							case channel <- text:
							case <-time.After(100 * time.Millisecond):
								t.logger.Warn().Str("text", text).Msg("failed to broadcast: channel blocked")
							}
						}
						return true
					})
				}
				recognizer.Reset(stream)
			}
		}
	}

}

curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/icefall-asr-zipformer-streaming-wenetspeech-20230615.tar.bz2

sherpa.OnlineRecognizerConfig{
		ModelConfig: sherpa.OnlineModelConfig{
			Transducer: sherpa.OnlineTransducerModelConfig{
				Encoder: "models/icefall-asr-zipformer-streaming-wenetspeech-20230615/exp/encoder-epoch-12-avg-4-chunk-16-left-128.onnx",
				Decoder: "models/icefall-asr-zipformer-streaming-wenetspeech-20230615/exp/decoder-epoch-12-avg-4-chunk-16-left-128.onnx",
				Joiner:  "models/icefall-asr-zipformer-streaming-wenetspeech-20230615/exp/joiner-epoch-12-avg-4-chunk-16-left-128.onnx",
			},
			Tokens:     "models/icefall-asr-zipformer-streaming-wenetspeech-20230615/data/lang_char/tokens.txt",
			NumThreads: 1,
			Debug:      0,
			ModelType:  "zipformer2",
			Provider:   "cpu",
		},
		DecodingMethod:          "greedy_search",
		MaxActivePaths:          4,
		EnableEndpoint:          1,
		Rule1MinTrailingSilence: 2.4,
		Rule2MinTrailingSilence: 1.2,
		Rule3MinUtteranceLength: 20,
		FeatConfig: sherpa.FeatureConfig{
			SampleRate: 16000,
			FeatureDim: 80,
		},
	}

  curl -SL -O https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20.tar.bz2

sherpa.OnlineRecognizerConfig{
		ModelConfig: sherpa.OnlineModelConfig{
			Transducer: sherpa.OnlineTransducerModelConfig{
				Encoder: "models/sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/encoder-epoch-99-avg-1.onnx",
				Decoder: "models/sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/decoder-epoch-99-avg-1.onnx",
				Joiner:  "models/sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/joiner-epoch-99-avg-1.onnx",
			},
			Tokens:       "models/sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/tokens.txt",
			NumThreads:   2,
			Debug:        0,
			Provider:     "cpu",
			ModelingUnit: "cjkchar",
			BpeVocab:     "models/sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/bpe.vocab",
		},

		HotwordsFile:            "models/sherpa-onnx-streaming-zipformer-bilingual-zh-en-2023-02-20/hotwords.txt",
		HotwordsScore:           2.0,
		HotwordsBufSize:         1024,
		DecodingMethod:          "modified_beam_search",
		MaxActivePaths:          4,
		EnableEndpoint:          1,
		Rule1MinTrailingSilence: 2.4,
		Rule2MinTrailingSilence: 1.2,
		Rule3MinUtteranceLength: 20,
		FeatConfig: sherpa.FeatureConfig{
			SampleRate: 16000,
			FeatureDim: 80,
		},
	}