Redditor builds a Star Trek-style home AI: the model was easy, the audio plumbing wasn't
FF2183 · reddit · 2026-10-27
A Reddit author details months of work building "Ada," an always-on Star Trek-style household voice AI, plus a new ESP32-based diagnostic panel called BM-01.
The stack
- Runs on a Lenovo ThinkCentre M720q Tiny: GPT-Live realtime voice, ReSpeaker XVF3800 four-mic array, local "Hey Ada" and "Red Alert" wake-word detection
- Persistent household memory, tool calling, WiZ lighting and Samsung TV control, speaker recognition, emergency states, a flight data recorder, and automatic recovery after reboots
Hard-won lessons
- The model wasn't the hard part — living with the system was: audio became a rabbit hole (AEC, VAD, interruption thresholds, and Patrick Stewart reliably triggering false wake-ups)
- Switching to the ReSpeaker's dedicated ASR USB audio channel gave an immediate improvement
- Barge-in failed because speaker playback blocked the async event loop; replacing it with a callback-driven audio queue fixed interruptions instantly
- A Chromium GPT-Live/WebRTC sandbox replaced the Realtime 2.1 audio stack, with night-and-day better conversation and interruption behavior
More from coding & agent
- Planning a Master-Agent Architecture: LangGraph vs Microsoft Agent Framework — DonXendon · 2026-10-27
- Score centering lands in TRL: default-on for async GRPO, stabilizes RL under mismatch — _lewtun · 2026-10-27
- Claude Code adds env var to skip even its last safety block, like rm -rf ~ — lydiahallie · 2026-10-27
- Codex ban update: users jailbreak AI agents to bypass Daybreak filter for reverse engineering — FlolightC · 2026-10-27
- GPT-6.1 Terra hits GA in GitHub Copilot, stronger terminal coding at lower cost than GPT-6.1 Sol — 0xkarasy · 2026-10-27
- MIT: LM harnesses are compositional generalizers, not just scaffolding — akbirthko · 2026-10-27