ESP32 voice assistant: AIMET AdaRound cuts wake-word model 8x without the 3.7% accuracy hit of naive 4-bit
carrycooldude · x · 2026-09-05
Developer @ashwinexe built a small voice assistant for the ESP32, inspired by a live on-device demo. Key details:
- Compressed the wake-word model 8x using Qualcomm's AIMET. Naive 4-bit rounding lost 3.7% accuracy; AIMET's AdaRound quantization recovered it.
- The wake word audio was generated with GradiumAI TTS in 22 different voices.
A practical reference for on-device quantization: naive 4-bit rounding genuinely hurts small models, and AdaRound-style methods are worth trying first.
More from Infra
- Tauri-Based Wan2GP Desktop Launcher Brings One-Click Open-Source Video Gen to Windows — Extension_Affect_483 · 2026-09-05
- MLX-DLSS: Hacker runs NVIDIA's DLSS 5 neural renderer on a Mac, within 0.005 MAE of the original — WaveCut · 2026-09-05
- Qwen3.8-27B on 2× RTX 5070 Ti: One vLLM Flag Kills Speed 11×, Full Benchmarks Inside — puthre · 2026-09-05
- Dev boosts GLM 5.2 TPS on a B300 and swaps it into Claude Code in place of Anthropic models — abhijithneil · 2026-09-05
- What forces LLM teams to optimize inference when going from MVP to production? — Ok_Philosophy_4031 · 2026-09-05
- Leaker claims GPT-6 Astra beats Claude Fable 5.1 at metaprompting — whoiskatrin · 2026-09-05