flutter/lib/audio/browser_voice_meter_web.dart

import 'dart:async';
import 'dart:js_interop';
import 'dart:math' as math;
import 'dart:typed_data';

import 'package:web/web.dart' as web;

class VoiceMeterFrame {
  const VoiceMeterFrame({required this.level, required this.samples});

  final double level;
  final List<double> samples;
}

class BrowserVoiceMeter {
  web.AudioContext? _context;
  web.AnalyserNode? _analyser;
  web.MediaStream? _stream;
  web.MediaStreamAudioSourceNode? _source;
  Timer? _timer;
  Float32List? _buffer;
  var _noiseFloor = 0.012;
  var _voiceCeiling = 0.08;
  var _smoothedLevel = 0.0;

  bool get isSupported => true;

  Future<void> start(void Function(VoiceMeterFrame frame) onFrame) async {
    await stop();

    final stream = await web.window.navigator.mediaDevices
        .getUserMedia(
          web.MediaStreamConstraints(
            audio: {
              'echoCancellation': true,
              'noiseSuppression': true,
              'autoGainControl': true,
              'channelCount': 1,
            }.jsify()!,
            video: false.toJS,
          ),
        )
        .toDart;
    final context = web.AudioContext();
    if (context.state == 'suspended') {
      await context.resume().toDart;
    }

    final source = context.createMediaStreamSource(stream);
    final analyser = context.createAnalyser()
      ..fftSize = 1024
      ..smoothingTimeConstant = 0.16;
    source.connect(analyser);

    _context = context;
    _stream = stream;
    _source = source;
    _analyser = analyser;
    _buffer = Float32List(analyser.fftSize);
    _noiseFloor = 0.012;
    _voiceCeiling = 0.08;
    _smoothedLevel = 0.0;

    _timer = Timer.periodic(const Duration(milliseconds: 45), (_) {
      final frame = _readFrame();
      onFrame(frame);
    });
  }

  Future<void> stop() async {
    _timer?.cancel();
    _timer = null;
    _source?.disconnect();
    _source = null;
    final stream = _stream;
    if (stream != null) {
      for (final track in stream.getTracks().toDart) {
        track.stop();
      }
    }
    _stream = null;
    _analyser = null;
    final context = _context;
    _context = null;
    if (context != null && context.state != 'closed') {
      await context.close().toDart;
    }
  }

  void dispose() {
    _timer?.cancel();
    _timer = null;
  }

  VoiceMeterFrame _readFrame() {
    final analyser = _analyser;
    final buffer = _buffer;
    if (analyser == null || buffer == null) {
      return const VoiceMeterFrame(level: 0, samples: []);
    }

    analyser.getFloatTimeDomainData(buffer.toJS);
    var sumSquares = 0.0;
    var peak = 0.0;
    for (final sample in buffer) {
      final centered = sample.abs();
      peak = math.max(peak, centered);
      sumSquares += centered * centered;
    }

    final rms = math.sqrt(sumSquares / buffer.length);
    final rawLevel = math.max(peak * 0.70, rms * 2.8);
    if (rawLevel < _noiseFloor) {
      _noiseFloor = _noiseFloor * 0.88 + rawLevel * 0.12;
    } else {
      _noiseFloor = _noiseFloor * 0.992 + rawLevel * 0.008;
    }
    if (rawLevel > _voiceCeiling) {
      _voiceCeiling = _voiceCeiling * 0.68 + rawLevel * 0.32;
    } else {
      _voiceCeiling = math.max(
        _noiseFloor + 0.035,
        _voiceCeiling * 0.992 + rawLevel * 0.008,
      );
    }

    final range = math.max(0.035, _voiceCeiling - _noiseFloor);
    final gated = ((rawLevel - _noiseFloor - 0.008) / range).clamp(0.0, 1.0);
    final level = math.pow(gated, 0.58).toDouble();
    _smoothedLevel +=
        (level - _smoothedLevel) * (level > _smoothedLevel ? 0.5 : 0.22);

    return VoiceMeterFrame(
      level: _smoothedLevel,
      samples: _bucketSamples(buffer, _noiseFloor, range),
    );
  }

  List<double> _bucketSamples(
    Float32List buffer,
    double noiseFloor,
    double range,
  ) {
    const buckets = 12;
    final bucketSize = (buffer.length / buckets).floor();
    return List<double>.generate(buckets, (bucket) {
      final start = bucket * bucketSize;
      final end = bucket == buckets - 1
          ? buffer.length
          : math.min(start + bucketSize, buffer.length);
      var peak = 0.0;
      var sumSquares = 0.0;
      for (var index = start; index < end; index++) {
        final sample = buffer[index].abs();
        peak = math.max(peak, sample);
        sumSquares += sample * sample;
      }
      final count = math.max(1, end - start);
      final rms = math.sqrt(sumSquares / count);
      final raw = math.max(peak * 0.75, rms * 2.8);
      final normalized = ((raw - noiseFloor) / range).clamp(0.0, 1.0);
      return (0.08 + math.pow(normalized, 0.58) * 0.92).clamp(0.0, 1.0);
    });
  }
}