Keyword Spotter API

Streaming keyword spotting API reference for sherpa-onnx-node.

Source file

scripts/node-addon-api/lib/keyword-spotter.js

API

KeywordSpotter

Streaming keyword spotter. Detects keywords in audio as it arrives.

Constructor

const spotter = new sherpa_onnx.KeywordSpotter(config);

param config

Configuration object with:

• 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 }

  • zipformerCtc — { model: string }

  • nemoCtc — { model: string }

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

• maxActivePaths (number, optional) — Max active paths for beam search.

• numTrailingBlanks (number, optional) — Number of trailing blanks.

• keywordsScore (number, optional) — Score boost for keywords.

• keywordsThreshold (number, optional) — Threshold for keyword detection.

• keywordsFile (string, optional) — Path to keywords file.

Methods

spotter.createStream()

returns

A new OnlineStream.

spotter.isReady(stream)

Check if the stream has enough data for decoding.

param stream

An OnlineStream.

returns

true if ready (boolean).

spotter.decode(stream)

Trigger one decoding step on the stream.

param stream

An OnlineStream.

spotter.reset(stream)

Reset the stream for a new utterance.

param stream

An OnlineStream.

spotter.getResult(stream)

Get the current keyword detection result.

param stream

An OnlineStream.

returns

A KeywordResult object with:

• keyword (string) — Detected keyword text.

• start_time (number) — Start time in seconds.

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

• tokens (string[]) — Token strings.

Properties

• spotter.config — The configuration object.

Example

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

const spotter = new sherpa_onnx.KeywordSpotter({
  modelConfig: {
    transducer: {
      encoder: './encoder.onnx',
      decoder: './decoder.onnx',
      joiner: './joiner.onnx',
    },
    tokens: './tokens.txt',
  },
  keywordsFile: './keywords.txt',
});

const stream = spotter.createStream();

// Feed audio incrementally
stream.acceptWaveform({ samples: audioChunk, sampleRate: 16000 });

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

const result = spotter.getResult(stream);
if (result.keyword) {
  console.log(`Detected: ${result.keyword}`);
}

spotter.reset(stream);

Notes

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

• Use keywordsFile to define custom keywords.

• Call reset() between utterances.