2 Audio Engine Deep Dive

2. Audio Engine Deep Dive

Tincan's audio engine (src/audio/) is built for sub-30ms real-time voice communications over lossy networks.

---

2.1 Audio Pipeline Architecture

 [ Microphone ]
       │ (cpal capture callback)
       ▼
 [ Lock-Free Ring Buffer (rtrb) ]
       │ (audio processing thread)
       ▼
 [ Voice Activity Detector (VAD) ]
       │ (speech detected)
       ▼
 [ Opus Encoder (audiopus) ]
       │ (encoded frames)
       ▼
 [ QUIC Datagram Network Mesh ]
       │ (incoming UDP packets)
       ▼
 [ Adaptive Jitter Buffer ]
       │ (reordered & PLC concealed)
       ▼
 [ Opus Decoder ]
       │ (PCM frames)
       ▼
 [ Multi-Source Mixer & Limiter ]
       │ (cpal playback callback)
       ▼
 [ Speakers ]

---

2.2 Core Modules

1. Audio Hardware Bridge (src/audio/device.rs)

• Uses cpal to interface with native system sound servers: CoreAudio (macOS), ALSA / PulseAudio / PipeWire (Linux), WASAPI (Windows).
• Enforces native 48000 Hz sample rate and 20ms frame lengths (960 samples per frame).

2. Lock-Free Thread Safety (rtrb)

• Audio hardware callbacks run in high-priority real-time threads.
• Real-time threads must never block, lock mutexes, allocate memory, or perform I/O.
• Tincan uses lock-free Single-Producer Single-Consumer (SPSC) ring buffers (rtrb) to bridge audio callbacks with network worker threads.

3. Opus Codec & Loss Concealment (src/audio/codec.rs)

• Encodes 16-bit PCM mono audio into 48 kHz Opus frames (typically 40–80 bytes per packet).
• Features built-in Packet Loss Concealment (PLC). When network packets drop, the decoder interpolates missing audio frames based on linear predictive coding.

4. Voice Activity Detection (src/audio/vad.rs)

• Calculates frame Root Mean Square (RMS) energy against background noise floors.
• Enables Discontinuous Transmission (DTX): suppresses packet generation during silence, cutting upload bandwidth to zero when not speaking.

5. Per-Peer Adaptive Jitter Buffer (src/audio/jitter.rs)

• Smooths out packet arrival jitter caused by Wi-Fi or internet routing variance.
• Reorders out-of-order sequence numbers and dynamically adjusts buffer depth to minimize latency while preventing dropouts.

6. Multi-Source Audio Mixer & Soft Limiter (src/audio/mixer.rs)

• Sums PCM samples from all active speakers in a channel.
• Employs a soft limiter to prevent digital clipping when multiple peers talk simultaneously.