Flutter Recorder Streaming

flutter_recorder audio streaming

flutter_recorder provides low-latency, real-time streaming of captured audio data via Dart streams. It supports streaming raw uncompressed PCM samples in any bit depth, as well as live-encoded Opus packets.

Minimal example

import 'dart:async';
import 'dart:typed_data';
import 'package:flutter_recorder/flutter_recorder.dart';

StreamSubscription<AudioDataContainer>? _audioSubscription;

Future<void> startStreaming() async {
  final recorder = Recorder.instance;

  await recorder.init(
    format: PCMFormat.s16le, // or f32le, s32le, etc.
    sampleRate: 16000,       // 16kHz common for Speech-to-Text
    channels: RecorderChannels.mono,
  );
  recorder.start();

  // 1. Listen to incoming audio data
  _audioSubscription = recorder.uint8ListStream.listen((AudioDataContainer data) {
    // IMPORTANT: Make a copy if buffering or processing asynchronously!
    final Uint8List frameCopy = Uint8List.fromList(data.rawData);

    // Send frameCopy over WebSocket or process it:
    // webSocket.add(frameCopy);
  });

  // 2. Start streaming PCM samples
  recorder.startStreamingData(format: StreamingFormat.pcm);
}

Future<void> stopStreaming() async {
  final recorder = Recorder.instance;
  recorder.stopStreamingData();
  await _audioSubscription?.cancel();
  recorder.stop();
}

Formats: PCM vs Opus Streaming

Configure the streaming mode in startStreamingData:

// Stream raw PCM bytes
Recorder.instance.startStreamingData(format: StreamingFormat.pcm);

// Stream length-prefixed Opus packets
Recorder.instance.startStreamingData(format: StreamingFormat.opus);

Opus Streaming Details

When streaming with StreamingFormat.opus:

• data.rawData contains length-prefixed Opus packets.
• Each packet begins with a 4-byte little-endian integer indicating the byte size of the ensuing Opus frame, followed by the encoded Opus payload.

Converting Sample Formats with AudioDataContainer

The emitted AudioDataContainer wraps the incoming byte buffer. You can access raw bytes via data.rawData or convert to specific Dart typed arrays:

recorder.uint8ListStream.listen((AudioDataContainer data) {
  // Access raw bytes (matches format passed to init()):
  final Uint8List raw = data.rawData;
  final int byteLength = data.length;

  // Convert to Float32 [-1.0, 1.0] (e.g. for ML models / visualizers):
  final Float32List floatSamples = data.toF32List(from: PCMFormat.s16le);

  // Convert to Int16 [-32768, 32767] (e.g. for standard PCM 16-bit encoders):
  final Int16List int16Samples = data.toS16List(from: PCMFormat.f32le);

  // Other available conversions:
  // data.toU8List(from: format);        // Uint8List (0..255)
  // data.toS8List(from: format);        // Int8List (-128..127)
  // data.toS24List(from: format);       // Int8List 3-byte aligned
  // data.toS24ListOnInt32(from: format);// Int32List 24-bit sign-extended
  // data.toS32List(from: format);       // Int32List (-2147483648..2147483647)
});

Tip: For maximum performance, initialize Recorder.instance.init(format: ...) directly with your target format to avoid costly per-frame conversions in Dart.

CRITICAL: Buffer Memory Reuse Rule

[!CAUTION] Audio data must be processed as it is received. To optimize performance and avoid high GC overhead, the native backend reuses the same memory buffer for successive stream events.

If you store frames in a list, queue them, or use reactive operators like RxDart.bufferTime(), all items in your list will point to the same overwritten buffer.

Solution: Always clone the buffer when retaining or delegating asynchronously:

final safeCopy = Uint8List.fromList(data.rawData);

Streaming to Speech-to-Text / AI APIs (e.g. Gemini Live, Whisper)

import 'dart:typed_data';
import 'package:flutter_recorder/flutter_recorder.dart';
import 'package:web_socket_channel/web_socket_channel.dart';

void setupSpeechStreaming(WebSocketChannel channel) {
  Recorder.instance.init(
    format: PCMFormat.s16le,
    sampleRate: 16000,
    channels: RecorderChannels.mono,
  );
  Recorder.instance.start();

  Recorder.instance.uint8ListStream.listen((data) {
    // Clone before sending to network sink
    final payload = Uint8List.fromList(data.rawData);
    channel.sink.add(payload);
  });

  Recorder.instance.startStreamingData(format: StreamingFormat.pcm);
}

The API Shape

• Stream<AudioDataContainer> get uint8ListStream: Broadcast stream of audio data packets.
• void startStreamingData({StreamingFormat format = StreamingFormat.pcm}): Begins streaming captured microphone data to uint8ListStream.
• void stopStreamingData(): Pauses emission to uint8ListStream.

Traps & Gotchas

• Do not forget to call startStreamingData(): Listening to uint8ListStream alone will not emit events until startStreamingData() is called.
• start() must be called before streaming: If the capture engine is not started, no audio packets will arrive.
• Conversion overhead: Performing software PCM conversion (toF32List, toS16List) on every 20ms audio frame on the main isolate can affect UI frames. Match the initialized PCMFormat to your required format whenever possible.
• Opus packets are length-prefixed: When parsing Opus streaming packets, read the 4-byte little-endian length prefix before parsing each payload.

Keeping this skill current

This skill ships inside the flutter_recorder package. To check for updates, run:

dart run flutter_recorder:skills --check