Supertonic
Lightning-fast on-device multilingual TTS — 99M params, 1000+ chars/sec on an M1 Mac, no cloud.
About
Supertonic by Supertone Inc. is a lightning-fast, fully on-device text-to-speech system. The 99M-parameter model synthesizes speech across 31 languages through ONNX Runtime with no cloud dependency — reaching 1000+ characters per second on an M1 Mac, fast enough to convert a full webpage to audio in under a second. It includes built-in text normalization, streaming synthesis, and a Python SDK plus local HTTP server with an OpenAI-compatible endpoint.
The honest picture: 99M parameters is deliberately tiny, so prosody and naturalness won't match ElevenLabs or the largest cloud models on difficult text — independent benchmarks are still emerging. The model is under OpenRAIL-M (not a fully permissive OSS license), which matters for some commercial uses. But for privacy-sensitive, offline, or latency-critical applications, the speed-to-quality ratio is exceptional.
What it can do
- On-device TTS in 31 languages
- 1000+ characters/second on Apple Silicon
- Streaming synthesis support
- Built-in text normalization
- OpenAI-compatible local HTTP server
- Flutter SDK for mobile
Setup
pip install supertonic, then from supertonic import TTS — the model downloads from Hugging Face automatically on first run.
Prerequisites
Python 3.9+; works on CPU via ONNX Runtime, no GPU required.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 53; Trending score 74/100 from researcher assessment.
★ 2,600 GitHub stars
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.