Show HN: Ito – streaming neural TTS for the ESP32-S3 (4.9 MB, no NPU, no cloud)

Category: ai-ml

Tags: text-to-speech, embedded, esp32, neural-networks, edge-computing

Score: 6.5/10 (Innovation: 8, Technical: 7, Documentation: 6, Utility: 5)

Ito is a streaming neural text-to-speech engine that runs entirely on an ESP32-S3 microcontroller without cloud or NPU, using a distilled 3.3M parameter model. It achieves real-time synthesis (RTF ~0.66) with 125ms first-chunk latency, demonstrating impressive optimization for resource-constrained embedded devices, though phonemization still runs on a host. The project is notable for its detailed measurements, board-specific verification, and potential to enable local voice interfaces on low-cost hardware.

Target audience: embedded developers, IoT engineers, makers

Repository: https://github.com/lokutor-ai/ito · C · GPL-3.0 · 29 stars

View on Hacker News