Speech Denoiser API

Speech denoising API reference for sherpa-onnx-node.

Source file

API

OfflineSpeechDenoiser

Removes background noise from a complete audio recording.

Constructor

const denoiser = new sherpa_onnx.OfflineSpeechDenoiser(config);
param config

Configuration object with:

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

    • gtcrn — { model: string } path to the GTCRN ONNX model.

    • dpdfnet — { model: string, attenuationLimitDb?: number }. The attenuation limit is offline-only, defaults to 0 (disabled), and accepts finite values in [0, 100]; infinity also disables it.

    Plus common fields: numThreads, debug, provider.

Methods

denoiser.run(obj)

Run denoising on the input audio.

param obj

Audio request object with:

  • samples (Float32Array) — Audio samples in [-1, 1].

  • sampleRate (number) — Sample rate in Hz.

  • enableExternalBuffer (boolean, optional, default true).

returns

A GeneratedAudio object with samples (Float32Array) and sampleRate (number).

Properties

  • denoiser.config — The configuration object.

  • denoiser.sampleRate — Expected input sample rate in Hz (number).

OnlineSpeechDenoiser

Removes background noise from audio in a streaming fashion.

Constructor

const denoiser = new sherpa_onnx.OnlineSpeechDenoiser(config);
param config

Configuration object with:

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

    • gtcrn — { model: string } path to the GTCRN ONNX model.

    • dpdfnet — { model: string } path to the DPDFNet ONNX model. All official 8, 16, and 48 kHz DPDFNet streaming exports are supported.

    Plus common fields: numThreads, debug, provider.

Methods

denoiser.run(obj)

Process a chunk of audio.

param obj

Audio request object with:

  • samples (Float32Array) — Audio samples in [-1, 1].

  • sampleRate (number) — Sample rate in Hz.

  • enableExternalBuffer (boolean, optional, default true).

returns

A GeneratedAudio object with samples (Float32Array) and sampleRate (number).

denoiser.flush(enableExternalBuffer?)

Flush remaining buffered audio and return denoised output.

param enableExternalBuffer

Whether to use an external buffer (boolean, default true).

returns

A GeneratedAudio object with samples (Float32Array) and sampleRate (number).

denoiser.reset()

Reset the internal state for reuse.

Properties

  • denoiser.config — The configuration object.

  • denoiser.sampleRate — Expected input sample rate in Hz (number).

  • denoiser.frameShiftInSamples — Frame shift in samples (number).

Example

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

// Offline denoising
const denoiser = new sherpa_onnx.OfflineSpeechDenoiser({
  model: { gtcrn: './gtcrn.onnx' },
});

const wave = sherpa_onnx.readWave('./noisy-audio.wav');
const denoised = denoiser.run({
  samples: wave.samples,
  sampleRate: wave.sampleRate,
});
sherpa_onnx.writeWave('clean-audio.wav', {
  samples: denoised.samples,
  sampleRate: denoised.sampleRate,
});

// Online denoising
const onlineDenoiser = new sherpa_onnx.OnlineSpeechDenoiser({
  model: { gtcrn: './gtcrn.onnx' },
});

// Process chunks as they arrive
const chunk1 = onlineDenoiser.run({ samples: audioChunk1, sampleRate: 16000 });
const chunk2 = onlineDenoiser.run({ samples: audioChunk2, sampleRate: 16000 });
const remaining = onlineDenoiser.flush();

onlineDenoiser.reset();  // ready for next audio

Notes

  • OfflineSpeechDenoiser processes the entire audio at once.

  • OnlineSpeechDenoiser processes audio incrementally; call flush() after the last chunk to get any remaining buffered audio.

  • Call reset() on the online denoiser to reuse it for new audio.

  • DPDFNet’s attenuationLimitDb is applied only by OfflineSpeechDenoiser. See Offline attenuation limit for its exact behavior.