MediaPipe Audio Task that performs automatic speech recognition (ASR) in unary and streaming modes via WASM SIMD or WebGPU.
Signature:
export declare class AudioTranscriber extends TaskRunner
Extends: TaskRunner
Methods
| Method | Modifiers | Description |
|---|---|---|
| close() | Closes the AudioTranscriber and releases native engine and Wasm memory. |
|
| createFromModelBuffer(wasmFileset, modelAssetBuffer) | static |
Initializes the Wasm runtime and creates a new AudioTranscriber from a model asset buffer or stream. |
| createFromModelPath(wasmFileset, modelAssetPath) | static |
Initializes the Wasm runtime and creates a new AudioTranscriber from a model asset path. |
| createFromOptions(wasmFileset, options) | static |
Initializes the Wasm runtime and creates a new AudioTranscriber based on the provided options. |
| finishStream() | Flushes any remaining buffered audio and finalizes the current stream. Finalized transcription segments. | |
| prewarm() | Prewarms the underlying inference engine to reduce first-chunk latency. | |
| setOptions(options) | Sets the options for the AudioTranscriber and initializes the underlying engine. Options cannot be updated once the engine has been created. |
|
| transcribe(audioSamples, endOfStream) | Transcribes mono PCM float audio samples in [-1.0, 1.0]. |
|
| transcribeStream(audioSamples) | Feeds a streaming chunk of mono PCM float audio samples (endOfStream = false) and returns any newly emitted interim or final segments. |
AudioTranscriber.close()
Closes the AudioTranscriber and releases native engine and Wasm memory.
Signature:
close(): void;
Returns:
void
AudioTranscriber.createFromModelBuffer()
Initializes the Wasm runtime and creates a new AudioTranscriber from a model asset buffer or stream.
Signature:
static createFromModelBuffer(wasmFileset: WasmFileset, modelAssetBuffer: Uint8Array | ReadableStreamDefaultReader): Promise<AudioTranscriber>;
Parameters
| Parameter | Type | Description |
|---|---|---|
| wasmFileset | WasmFileset | A configuration object that provides the location of the Wasm binary and its loader. |
| modelAssetBuffer | Uint8Array | ReadableStreamDefaultReader | A Uint8Array or ReadableStreamDefaultReader containing the .litertlm model bytes. |
Returns:
Promise<AudioTranscriber>
AudioTranscriber.createFromModelPath()
Initializes the Wasm runtime and creates a new AudioTranscriber from a model asset path.
Signature:
static createFromModelPath(wasmFileset: WasmFileset, modelAssetPath: string): Promise<AudioTranscriber>;
Parameters
| Parameter | Type | Description |
|---|---|---|
| wasmFileset | WasmFileset | A configuration object that provides the location of the Wasm binary and its loader. |
| modelAssetPath | string | The path or URL to the .litertlm model asset. |
Returns:
Promise<AudioTranscriber>
AudioTranscriber.createFromOptions()
Initializes the Wasm runtime and creates a new AudioTranscriber based on the provided options.
Signature:
static createFromOptions(wasmFileset: WasmFileset, options: AudioTranscriberOptions): Promise<AudioTranscriber>;
Parameters
| Parameter | Type | Description |
|---|---|---|
| wasmFileset | WasmFileset | A configuration object that provides the location of the Wasm binary and its loader. |
| options | AudioTranscriberOptions | The options for the AudioTranscriber. Note that either a path to the model or the model buffer itself must be provided via baseOptions. |
Returns:
Promise<AudioTranscriber>
AudioTranscriber.finishStream()
Flushes any remaining buffered audio and finalizes the current stream.
Finalized transcription segments.
Signature:
finishStream(): Promise<AudioTranscriberResult>;
Returns:
Promise<AudioTranscriberResult>
AudioTranscriber.prewarm()
Prewarms the underlying inference engine to reduce first-chunk latency.
Signature:
prewarm(): Promise<void>;
Returns:
Promise<void>
AudioTranscriber.setOptions()
Sets the options for the AudioTranscriber and initializes the underlying engine. Options cannot be updated once the engine has been created.
Signature:
setOptions(options: AudioTranscriberOptions): Promise<void>;
Parameters
| Parameter | Type | Description |
|---|---|---|
| options | AudioTranscriberOptions | The options for the AudioTranscriber. |
Returns:
Promise<void>
AudioTranscriber.transcribe()
Transcribes mono PCM float audio samples in [-1.0, 1.0].
Signature:
transcribe(audioSamples: Float32Array | number[], endOfStream?: boolean): Promise<AudioTranscriberResult>;
Parameters
| Parameter | Type | Description |
|---|---|---|
| audioSamples | Float32Array | number[] | Mono PCM float audio samples at sampleRateHz. |
| endOfStream | boolean | Whether this chunk marks the end of the audio stream. Defaults to true (unary single-utterance mode). The transcription result containing segments and concatenated text. |
Returns:
Promise<AudioTranscriberResult>
AudioTranscriber.transcribeStream()
Feeds a streaming chunk of mono PCM float audio samples (endOfStream = false) and returns any newly emitted interim or final segments.
Signature:
transcribeStream(audioSamples: Float32Array | number[]): Promise<AudioTranscriberResult>;
Parameters
| Parameter | Type | Description |
|---|---|---|
| audioSamples | Float32Array | number[] | Mono PCM float audio samples at sampleRateHz. Newly emitted segments from the streaming session. |
Returns:
Promise<AudioTranscriberResult>