Major rewrite of the voice assistant’s audio processing pipeline, delivering crystal-clear 24kHz native audio playback and eliminating clicks, pops, and timing drift in long responses.
STATUS [v0.9.2]
Problems Solved
| Issue | Cause | Solution |
- | Muffled voice | 24kHz→16kHz downsampling | Dual AudioContext architecture |
- | Exact ±1.0 peaks | Implicit endianness | DataView with explicit little-endian |
- | Clicks/pops | Queue-based callback latency | Proactive scheduling |
- | Timing drift | Cumulative callback delays | Immediate chunk scheduling |
Architecture Improvements
- Dual AudioContext: Separate contexts for capture (16kHz) and playback (24kHz)
- Proactive Scheduling: Chunks scheduled immediately with precise timing, no queue latency
- Explicit Endianness: DataView ensures correct little-endian PCM handling per Google’s recommendation
- Master Gain Control: Single gain node for consistent volume across all chunks