Class Transcriber

constructor

• new Transcriber(
      modelURL: string,
      callbacks?: Partial<TranscriberCallbacks>,
      useVAD?: boolean,
      precision?: string,
  ): Transcriber

  Creates a transcriber for transcribing a MediaStream from any source. After creating the Transcriber, you must invoke Transcriber.attachStream to provide a MediaStream that you want to transcribe.

  Parameters

  • modelURL: string

    The URL that the underlying MoonshineModel weights should be loaded from, relative to Settings.BASE_ASSET_PATH.MOONSHINE.
  • callbacks: Partial<TranscriberCallbacks> = {}

    A set of TranscriberCallbacks used to trigger behavior at different steps of the transcription lifecycle. For transcription-only use cases, you should define the TranscriberCallbacks yourself; when using the transcriber for voice control, you should create a VoiceController and pass it in.
  • useVAD: boolean = true

    A boolean specifying whether or not to use Voice Activity Detection (VAD) for deciding when to perform transcriptions. When set to true, the transcriber will only process speech at the end of each chunk of voice activity; when set to false, the transcriber will operate in streaming mode, generating continuous transcriptions on a rapid interval.
  • precision: string = "quantized"

  Returns Transcriber

  Example

  This basic example demonstrates the use of the transcriber with custom callbacks:

  import Transcriber from "@moonshine-ai/moonshine-js";
  
  var transcriber = new Transcriber(
       "model/tiny",
       {
           onModelLoadStarted() {
               console.log("onModelLoadStarted()");
           },
           onTranscribeStarted() {
               console.log("onTranscribeStarted()");
           },
           onTranscribeStopped() {
               console.log("onTranscribeStopped()");
           },
           onTranscriptionUpdated(text: string | undefined) {
               console.log(
                   "onTranscriptionUpdated(" + text + ")"
               );
           },
           onTranscriptionCommitted(text: string | undefined) {
               console.log(
                   "onTranscriptionCommitted(" + text + ")"
               );
           },
       },
       false // use streaming mode
  );
  
  // Get a MediaStream from somewhere (user mic, active tab, an <audio> element, WebRTC source, etc.)
  ...
  
  transcriber.attachStream(stream);
  transcriber.start();
  Copy

callbacks

isActive

attachStream

• attachStream(stream: MediaStream): void

  Attaches a MediaStream to this Transcriber for transcription. A MediaStream must be attached before starting transcription.

  Parameters

  • stream: MediaStream

    A MediaStream to transcribe

  Returns void

detachStream

• detachStream(): void

  Detaches the MediaStream used for transcription. TODO

  Returns void

getAudioBuffer

• getAudioBuffer(buffer: Float32Array): AudioBuffer

  Returns the most recent AudioBuffer that was input to the underlying model for text generation. This is useful in cases where we want to double-check the audio being input to the model while debugging.

  Parameters

  • buffer: Float32Array

  Returns AudioBuffer

  An AudioBuffer

load

• load(): Promise<void>

  Preloads the models and initializes the buffer required for transcription.

  Returns Promise<void>

start

• start(): Promise<void>

  Starts transcription.

  Transcription will stop when stop is called.

  Note that the Transcriber must have a MediaStream attached via Transcriber.attachStream before starting transcription.

  Returns Promise<void>

stop

• stop(): void

  Stops transcription.

  Returns void