前端音频处理架构

graph TB
    subgraph capture [音频采集]
        MIC["麦克风\ngetUserMedia()"]
        CTX["AudioContext\n(16kHz)"]
        WL["AudioWorkletNode\ncapture-processor"]
        MIC --> CTX --> WL
    end

    subgraph transport [传输]
        WS["WebSocket\naudio_base64"]
    end

    subgraph playback [音频播放]
        AP["AudioPlayer"]
        RESAMP["重采样\n24kHz → 设备采样率"]
        BUF["AudioBufferSourceNode\n预调度播放"]
        AP --> RESAMP --> BUF
    end

    subgraph analysis [音频分析]
        LUFS["LUFS 测量\nITU-R BS.1770"]
        MIXER["MixerController\n自动增益"]
        LUFS --> MIXER
    end

    subgraph recording [录制]
        SREC["SessionRecorder\n立体声 WAV"]
        VREC["SessionVideoRecorder\n视频 + 音频"]
    end

    WL -->|"Float32 chunk (1s)"| WS
    WS -->|"Base64 audio (24kHz)"| AP
    AP --> SREC
    AP --> VREC
    BUF --> LUFS

process(inputs, outputs) {
    // 1. 直通到输出（用于 MediaStreamDestination）
    output.set(input);
    // 2. 累积到 _buffer
    _buffer = concat(_buffer, input);
    // 3. buffer 满一个 chunk 时发送
    while (_buffer.length >= _chunkSize) {
        const chunk = _buffer.slice(0, _chunkSize);
        _buffer = _buffer.slice(_chunkSize);
        port.postMessage({type: 'chunk', audio: chunk}, [chunk.buffer]);
    }
}

sequenceDiagram
    participant S as 服务端
    participant AP as AudioPlayer
    participant ACX as AudioContext

    S->>AP: playChunk(base64)
    AP->>AP: Base64 → Float32Array
    AP->>AP: 重采样 24kHz → 设备采样率
    AP->>AP: 创建 AudioBuffer
    
    alt 首个 chunk + 配置了延迟
        AP->>AP: setTimeout(playbackDelay)
        Note over AP: 等待更多 chunk 到达避免间隙
    end
    
    AP->>ACX: bufferSource.start(nextTime)
    AP->>AP: nextTime += buffer.duration
    
    Note over AP: 后续 chunk 紧密拼接
    S->>AP: playChunk(base64)
    AP->>ACX: bufferSource.start(nextTime)

gapMs = (currentTime - nextTime) * 1000
if gapMs > 10ms:
    gapCount++
    totalShiftMs += gapMs
    nextTime = currentTime + 小偏移    // 修正
    触发 onGap 回调