Supertonic 3: 99M-Parameter On-Device TTS Across 31 Languages, No GPU Required

Supertone’s Supertonic 3 is a 99M-parameter open-weight text-to-speech model that runs on CPU, supports 31 languages, and outputs 44.1kHz audio on devices as small as a Raspberry Pi.
artificial-intelligence
Author

Kabui, Charles

Published

2026-05-28

Keywords

text-to-speech, on-device-ai, supertonic, multilingual-tts, onnx-runtime

Download as PDF

Supertone released Supertonic 3, a 99-million-parameter text-to-speech model that runs entirely on-device through ONNX Runtime (a cross-platform inference engine). It supports 31 languages from a single checkpoint, outputs audio at 44.1kHz, and needs no GPU. The model runs on desktop, mobile, Raspberry Pi, and even e-readers in airplane mode. Ten inline expression tags (<laugh>, <breath>, <sigh>) add human nuance without reference audio. A language-agnostic mode auto-detects input language when unspecified. SDKs cover 11 platforms including Python, Node.js, WebGPU browsers, Java, C++, Swift, Rust, and Flutter. The Python SDK includes a local HTTP server with an OpenAI-compatible /v1/audio/speech endpoint, so existing apps can switch by changing a URL. Weights are open under the OpenRAIL-M license.

Cloud TTS services charge per character and require an internet connection. Supertonic 3 runs offline, costs nothing after download, and fits in 99M parameters, a fraction of the 0.7B to 2B range typical of competing open TTS systems. A developer building a voice assistant can run pip install supertonic and generate speech locally in under a second. An e-reader can narrate books with zero network dependency. Any tool already using the OpenAI TTS API can switch to Supertonic’s local server with one URL change.

On-device audio AI has closed the gap with cloud services faster than most expected. When a 99M-parameter model produces usable quality across 31 languages on a $35 Raspberry Pi, the argument for paying per-character API fees weakens for most use cases outside premium voice cloning. For more on the broader on-device AI trend, see Unsloth Studio + Qwen3.6: Running Frontier AI on 18GB RAM.

Sources:

Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service

Reuse

GNU GENERAL PUBLIC LICENSE v3.0(View License)

Citation

BibTeX citation:
@misc{kabui2026,
  author = {{Kabui, Charles}},
  title = {Supertonic 3: {99M-Parameter} {On-Device} {TTS} {Across} 31
    {Languages,} {No} {GPU} {Required}},
  date = {2026-05-28},
  url = {https://toknow.ai/posts/supertonic-3-on-device-tts-99m-parameters-31-languages/},
  langid = {en-GB}
}
For attribution, please cite this work as:
Kabui, Charles. 2026. “Supertonic 3: 99M-Parameter On-Device TTS Across 31 Languages, No GPU Required.” https://toknow.ai/posts/supertonic-3-on-device-tts-99m-parameters-31-languages/.