sanoTTS: a 337KB neural TTS that runs on a $3 ESP32-S3 or in your browser
alexcovo_eth · x · 2026-09-08
Open-source sanoTTS pushes neural TTS to the extreme: a model family of just 294k–2.3M parameters, with the smallest weighing only 337KB.
Key points:
- Runs in real time on a $3 ESP32-S3 microcontroller — no cloud, no NPU; wire a GPIO into an LM386 and speaker
- Runs directly in the browser via WebAssembly, no text leaves the device
- Under 4MB per voice, zero dependencies (espeak-ng phonemizer included)
- 11 voices across 6 languages (English, Nepali, Hindi, Vietnamese, Indonesian, Chinese)
- Leads SCOREQ/UTMOS in the sub-15M parameter class
- Available via Python, NPM, and Arduino/PlatformIO
The author argues that squeezing AI into 337KB — letting toys and MCUs speak on their own — is more interesting than making models ever bigger. The repo has 400 stars on GitHub.
More from Infra
- Extrapolating OpenAI's plots: agents may eat 25% of compute within 9-12 months — joshua_clymer · 2026-09-08
- User runs Pinokio with Wan2GP, MiniMax H3, Viggle-Animate on DGX Spark — cocktailpeanut · 2026-09-08
- Lerna plugin routes GitHub Copilot HydraFusion model calls to Azure AI Foundry — unixterminal · 2026-09-08
- Reactor ships MiniMax FastH3 with NVIDIA SANA, first video generation 3x faster than realtime — flngr · 2026-09-08
- NVIDIA's new Sol-H3 fast inference method for H3 awaits a ComfyUI port — krigeta1 · 2026-09-08
- Walking the AI rack optical stack: InP substrates and silicon photonics as the cleaner bet — demian_ai · 2026-09-08