@cantoo/capacitor-audio-capture

Capacitor v7 plugin that captures microphone audio on web / Android / iOS and streams raw PCM as Float32Array (mono) chunks to the consumer. No format conversion, no compression — your app decides what to do with the samples.

Platforms

Native captures mono audio, resamples it to targetSampleRate via linear interpolation, applies optional RMS silence filtering, and ships each chunk over the bridge as base64 little-endian Float32 PCM. The JS wrapper decodes that base64 once and hands the consumer a Float32Array — no further encoding/decoding inside the plugin.

Installation

pnpm add @cantoo/capacitor-audio-capture
# or: npm i @cantoo/capacitor-audio-capture
# or: yarn add @cantoo/capacitor-audio-capture

npx cap sync

Permissions

• iOS — add to Info.plist:

  <key>NSMicrophoneUsageDescription</key>
  <string>Used to record audio while you use the app.</string>
  

• Android — RECORD_AUDIO is already declared by the plugin; permission is requested at runtime on the first call to startCapture.
• Web — requires a secure context (HTTPS or localhost) and the user must grant microphone access.

Usage

import { AudioCapture } from '@cantoo/capacitor-audio-capture';

await AudioCapture.startCapture({
  chunkDurationMs: 250,
  targetSampleRate: 16000,
  silenceThreshold: 0.01,
  listener: (sequence, chunk) => {
    // chunk: Float32Array, mono, at targetSampleRate (samples in -1..1)
    console.log(sequence, chunk.length);
  },
});

// later
await AudioCapture.stopCapture();
await AudioCapture.release();

Examples

Send to a backend (base64-encode just before the request):

listener: (_seq, chunk) => {
  const bytes = new Uint8Array(chunk.buffer);
  let bin = '';
  for (let i = 0; i < bytes.length; i++) bin += String.fromCharCode(bytes[i]);
  const base64 = btoa(bin);
  fetch('/api/audio', {
    method: 'POST',
    headers: { 'Content-Type': 'application/octet-stream' },
    body: chunk.buffer,
  });
}

Buffer for a client-side WAV:

await AudioCapture.startCapture({
  chunkDurationMs: 250,
  targetSampleRate: 16000,
  silenceThreshold: 0,
  accumulate: true,
});
// ... user stops ...
const { audio } = await AudioCapture.stopCapture();
// `audio` is a Float32Array of the full session — convert to Int16,
// prepend a 44-byte WAV header, and you have a playable file.

API

startCapture(options?)

sequence is monotonically increasing per session and resets on every startCapture. It is incremented even for chunks dropped by silence detection, so consumers can detect temporal gaps.

stopCapture()

Stops capture and releases the active capture session. Returns { audio: Float32Array }. When the session was started with accumulate: true, audio is the concatenated PCM (mono Float32 at targetSampleRate); otherwise it is an empty Float32Array.

release()

Stops capture, removes listeners, and frees every open resource.

Exported types

import type {
  AudioCapturePlugin,
  AudioChunkListener,  // (sequence, chunk: Float32Array) => void
  AudioCaptureError,
  AudioCaptureErrorCode,
  StartCaptureOptions,
} from '@cantoo/capacitor-audio-capture';

Error handling

Every rejected Promise from the plugin carries a .code from a closed set, identical across iOS, Android, and Web. Branch on .code rather than parsing messages — the messages are short and stable, but only the code is part of the public contract.

import { AudioCapture, type AudioCaptureError } from '@cantoo/capacitor-audio-capture';

try {
  await AudioCapture.startCapture({ /* ... */ });
} catch (e) {
  const code = (e as AudioCaptureError).code;
  if (code === 'PERMISSION_DENIED') {
    // prompt user to enable mic access in OS settings
  } else if (code === 'MICROPHONE_UNAVAILABLE') {
    // surface a "no microphone" state in UI
  } else if (code === 'ALREADY_CAPTURING') {
    // ignore or call stopCapture() first
  }
}

Why base64 on the wire, Float32Array in the API?

Capacitor's native↔JS bridge can only carry JSON-serializable values. Encoding the PCM samples as base64 is ~33% smaller than a JSON number array and significantly faster to serialize/deserialize on both sides. The plugin's wrapper does the one decode (base64 → Float32Array) so the consumer always receives the natural representation for audio: a typed array of mono samples in [-1, 1].

Web and Content-Security-Policy

On web, the AudioWorkletProcessor is inlined as a string and loaded via URL.createObjectURL(new Blob([...])). If the host app has a strict CSP that doesn't allow blob: in script-src / worker-src, the worklet won't load — simply allow blob: in the appropriate directive.

License

MIT © Cantoo