Streaming ASR API

Streaming (online) speech recognition API reference for sherpa-onnx-node.

Source file

scripts/node-addon-api/lib/streaming-asr.js

API

OnlineRecognizer

Streaming speech recognizer. Processes audio incrementally and returns partial results as audio arrives.

Constructor

const recognizer = new sherpa_onnx.OnlineRecognizer(config);
param config

Recognizer configuration object (OnlineRecognizerConfig).

The config object supports:

  • featConfig (object, optional) — { sampleRate: number, featureDim: number }

  • modelConfig (object, optional) — Model configuration with one of:

    • transducer — { encoder: string, decoder: string, joiner: string }

    • paraformer — { encoder: string, decoder: string }

    • zipformer2Ctc — { model: string }

    • nemoCtc — { model: string }

    Plus common fields: tokens, numThreads, debug, provider.

  • decodingMethod (string, optional) — e.g. 'greedy_search'.

  • maxActivePaths (number, optional) — For beam search.

  • enableEndpoint (boolean, optional) — Enable endpoint detection.

  • rule1MinTrailingSilence (number, optional) — Endpoint rule 1.

  • rule2MinTrailingSilence (number, optional) — Endpoint rule 2.

  • rule3MinUtteranceLength (number, optional) — Endpoint rule 3.

  • blankPenalty (number, optional) — Blank penalty for CTC models.

Methods

recognizer.createStream()
returns

A new OnlineStream.

recognizer.isReady(stream)

Check if the stream has enough frames for decoding.

param stream

An OnlineStream.

returns

true if ready (boolean).

recognizer.decode(stream)

Trigger one decoding step on the stream.

param stream

An OnlineStream.

recognizer.isEndpoint(stream)

Check if an endpoint has been detected.

param stream

An OnlineStream.

returns

true if endpoint detected (boolean).

recognizer.reset(stream)

Reset the stream for a new utterance.

param stream

An OnlineStream.

recognizer.getResult(stream)

Get the current recognition result.

param stream

An OnlineStream.

returns

An OnlineRecognizerResult object with:

  • text (string) — Recognized text.

  • tokens (string[]) — Token strings.

  • timestamps (number[]) — Per-token timestamps.

  • is_final (boolean) — Whether this is a final result.

Properties

  • recognizer.config — The configuration object.

OnlineStream

An active streaming recognition stream.

Methods

stream.acceptWaveform(obj)

Feed audio to the stream.

param obj

{ samples: Float32Array, sampleRate: number }.

stream.inputFinished()

Signal that no more audio will be fed to this stream.

Display

Helper for printing recognized words to the console.

Constructor

const display = new sherpa_onnx.Display(maxWordPerLine);
param maxWordPerLine

Max words per line (number).

Methods

display.print(idx, text)

Print recognized text.

param idx

Segment index (number).

param text

Text to display (string).

Example

const sherpa_onnx = require('sherpa-onnx-node');

const recognizer = new sherpa_onnx.OnlineRecognizer({ /* config */ });
const stream = recognizer.createStream();

stream.acceptWaveform({ samples: audioSamples, sampleRate: 16000 });

while (recognizer.isReady(stream)) {
  recognizer.decode(stream);
}

const result = recognizer.getResult(stream);
console.log(result.text);

if (recognizer.isEndpoint(stream)) {
  recognizer.reset(stream);  // ready for next utterance
}

Notes

  • Feed audio continuously; call decode() whenever isReady() returns true.

  • Use isEndpoint() to detect utterance boundaries.

  • Call inputFinished() when no more audio will arrive.