Supertonic 3

Spec sheet

Purpose

Supertonic 3 is a 99M-parameter open-weight text-to-speech model that runs locally on CPU through ONNX Runtime.Read 12 Oct 2026 · supertonic3.github.io

Product

Supertonic 3 is a 99M-parameter open-weight multilingual text-to-speech model that runs locally on CPU using ONNX Runtime.Read 12 Oct 2026 · supertonic3.github.io

Platforms

web,ios,android,apiRead 20 Sep 2026 · supertonic3.github.io

Free plan

YesRead 20 Sep 2026 · supertonic3.github.io

API

The maker links a Supertone API for expressive voice generation across 31 languages with adjustable speech controls.Read 12 Oct 2026 · supertonic3.github.io

Integrations

The maker lists Python, Node.js, web, iOS, Android, and C++ runtimes with examples.Read 12 Oct 2026 · supertonic3.github.io

Headquarters

Seoul, Republic of KoreaRead 28 Sep 2026 · supertonic3.github.io

API access

YesRead 20 Sep 2026 · supertonic3.github.io

API service

The maker describes a Supertone API for expressive voice generation across 31 languages with adjustable speech controls.Read 12 Oct 2026 · supertonic3.github.io

Browser service

The linked Supertone Play page is described as offering 263 voices across scenes and emotions in a browser without installation.Read 12 Oct 2026 · supertonic3.github.io

Code license

The Python package, runtime bindings, and example code are MIT-licensed with attribution required.Read 12 Oct 2026 · supertonic3.github.io

Commercial use

The model weights use OpenRAIL-M, which allows commercial use subject to use restrictions and attribution requirements.Read 12 Oct 2026 · supertonic3.github.io

Developer runtimes

The maker lists Python, Node.js, web, iOS, Android, and C++ examples and install instructions.Read 12 Oct 2026 · supertonic3.github.io

Export formats

WAVRead 20 Sep 2026 · supertonic3.github.io

Languages

The model supports 31 languages without per-language fine-tuning.Read 12 Oct 2026 · supertonic3.github.io

Licensing limit

The maker states that OpenRAIL-M has downstream use restrictions and is not equivalent to MIT.Read 12 Oct 2026 · supertonic3.github.io

Local use

The maker says the model runs in a browser or fully offline on a machine, without a GPU, cloud service, or API.Read 12 Oct 2026 · supertonic3.github.io

Model license

The model weights use OpenRAIL-M, which the maker says permits commercial use subject to use-based restrictions and attribution.Read 12 Oct 2026 · supertonic3.github.io

Model limitation

The archived repository says its open-weight release focuses on fixed-voice local TTS and does not include an official voice-cloning pipeline.Read 12 Oct 2026 · github.com

Offline use

The model can run in a browser or fully offline on a local machine without a GPU, cloud connection, or API for synthesis.Read 12 Oct 2026 · supertonic3.github.io

Output

The archived repository says the model outputs 44.1 kHz, 16-bit WAV audio.Read 12 Oct 2026 · github.com

Paid service mention

The maker says a Pro Play Desktop subscription unlocks commercially licensed voices, zero-shot cloning, and unlimited usage.Read 12 Oct 2026 · supertonic3.github.io

Performance claim

In the maker's benchmark, Supertonic 3 produced audio at 5.00 times real time on a 16-thread CPU across 30 samples.Read 12 Oct 2026 · supertonic3.github.io

Preset voices

Supertone Play offers 263 voices across scenes and emotions and lets users try Supertonic 3 in a browser without installation.Read 12 Oct 2026 · supertonic3.github.io

Speech controls

The model supports expression tags such as <laugh>, <breath>, and <sigh>.Read 12 Oct 2026 · huggingface.co

Support status

The GitHub repository page says it was archived on September 9, 2026, and that development and support have ended, with no updates, bug fixes, security patches, or monitored issues and pull requests.Read 12 Oct 2026 · github.com

Voice cloning

The linked Voice Builder supports zero-shot cloning by recording or uploading a short reference and synthesizing across 31 languages.Read 12 Oct 2026 · supertonic3.github.io

Voice styles

The open-weight package includes fixed preset voice styles, while a linked Voice Builder offers zero-shot custom voice-style creation from reference audio.Read 12 Oct 2026 · huggingface.co

Where Specifiction places it

• Text-to-speech tools#120 of 2337.9
• Text-to-Speech Software#164 of 2237.5