Lokutor releases ItoTTS, a 4.89 MB streaming TTS engine for ESP32-S3

Lokutor has released ItoTTS, a streaming text-to-speech engine designed for the ESP32-S3 microcontroller. The project aims to enable local LLMs to have a voice on hardware costing approximately $5.

Performance and Specs

The engine supports two English voices and outputs 24 kHz audio. Each voice requires 4.89 MB of weights. In automatic naturalness evaluations using UTMOS scores on eight held-out sentences, ItoTTS achieved a score of 4.46, outperforming other ESP32-compatible models:

• Teacher (StyleTTS 2): 4.49
• Ito: 4.46
• sanoTTS amy: 3.98
• sanoTTS heart-nano: 2.07

The developers note that this is a small automatic evaluation, not an independent listening study. Current limitations include text-to-phoneme conversion running on the host rather than the device, and speed has not yet been measured on physical hardware.

Licensing and Availability

The code is open-source under GPLv3. Model weights are available on Hugging Face under CC BY-NC-SA 4.0 plus additional terms, meaning they are free for non-commercial use but require a separate written agreement for commercial applications. The team states this licensing choice is intended to protect against corporate exploitation while remaining open to collaboration with individuals and small companies.