PCM stream

While recording, the recorder can send you the raw audio in small chunks, for example to run speech recognition or voice activity detection while the user is still talking.

import { WaveformRecorderView } from 'react-native-waveform-recorder';
import {
  decodePcmChunk,
  pcmToMonoFloat32,
} from 'react-native-waveform-recorder/pcm-stream';

<WaveformRecorderView
  ref={ref}
  style={{ height: 56 }}
  output={{ format: 'wav', sampleRate: 16000, channels: 1 }}
  enablePcmStream
  pcmChunkMs={100}
  onPcmChunk={(e) => {
    const int16 = decodePcmChunk(e.chunk);
    const float32 = pcmToMonoFloat32(int16, e.channels);
    vad.push(float32, e.sampleRate);
  }}
/>;

Requirements

• output.format must be 'wav'. With any other format enablePcmStream is ignored and no chunk is sent.
• Set enablePcmStream before start(). The setting is read when a segment starts recording.
• The recording file is still written as usual and onComplete still fires.

The chunk

Field	Meaning
chunk	Base64 string of interleaved little-endian 16-bit PCM.
sampleRate	Sample rate in Hz, from output.sampleRate.
channels	Channel count.
bytesPerSample	Always 2.
timestampMs	Recorded time when the chunk was sent, in ms.

pcmChunkMs (default 200, minimum 20) sets the approximate chunk length:

• iOS reads new bytes from the WAV file every pcmChunkMs, at most 256 KB per chunk.
• Android collects samples from AudioRecord until it has about pcmChunkMs of audio, then sends them. The remaining samples are sent when recording stops.

Helpers

The helpers live in the react-native-waveform-recorder/pcm-stream subpath, so apps that do not import them do not ship them.

Function	What it does
decodePcmChunk(chunk)	Decodes the base64 payload into an Int16Array. Uses globalThis.atob when available and a built-in decoder otherwise.
pcmToMonoFloat32(int16, channels)	Converts interleaved Int16 into mono Float32Array in [-1, 1]. Stereo pairs are averaged.

See the API reference for signatures.

Limits

• Chunks cross the bridge as base64 strings. That is fine for speech at 16 kHz mono (32 KB per second), but for several MB per second consider a JSI audio library such as react-native-audio-api.
• On iOS, after pause() and resume() the stream starts reading the current file from the beginning again, so audio sent before the pause can be sent a second time. On iOS the last partial chunk before pause() or stop() is not sent. If exact continuity matters, use the final WAV file from onComplete as the source of truth.
• Chunks are delivered on the JavaScript thread. Keep onPcmChunk light and hand heavy work to a worker or native module.