Supertonic — Public project reference¶

Development line: project:supertonic · thread public-project-reference
Last event: 2026-05-18 · 1 dated since 2026-05-18 · Researched: 2026-09-05 · confidence: high

What it is¶

Supertonic — an ONNX Runtime text-to-speech stack for developers who need a self-hosted alternative to hosted TTS APIs.

• Python SDK, CLI, and loopback HTTP server
• Supertonic-3 with 31 language codes plus na fallback
• ten bundled voice styles and Voice Builder JSON import

Development line¶

• 2026-05-18 — Supertonic public repository reference. On 2026-05-18, the Supertonic thread linked to the project's public GitHub repository alongside a source reference. This establishes a dated public reference point for the project without asserting a specific release, feature, or code change.

What changed¶

• 2025-12-10 — the Python package and six additional built-in voice styles landed.
• 2026-01-06 — Supertonic 2 expanded the model from English to five languages; its code path stayed on a release branch.
• 2026-01-22 — Voice Builder went live for persistent custom voice profiles.
• 2026-04-29 — Supertonic 3 released with 31-language support, reading-stability improvements, and a v2-compatible public ONNX interface.
• 2026-05-18 — Python SDK 1.3.1 added the local supertonic serve HTTP interface; Voice Builder added Supertonic-3 JSON profiles.
• 2026-05-20 — Supertonic 3 became available in Supertone Play and the Supertone API.

How to use this¶

Start with the linked public GitHub repository as of 2026-05-18. We have no evidence for version-specific adoption guidance.

1. Install supertonic; use Python 3.9 or later. — https://pypi.org/project/supertonic/
2. Create TTS(auto_download=True), select a bundled voice, synthesize text, and save the WAV output. — https://supertone-inc.github.io/supertonic-py/quickstart/
3. Pass a supported ISO language code when known, or na for language-agnostic handling. — https://supertone-inc.github.io/supertonic-py/quickstart/
4. For HTTP clients, install supertonic[serve], run supertonic serve --host 127.0.0.1 --port 7788, then call /v1/tts or /v1/audio/speech. — https://supertone-inc.github.io/supertonic-py/quickstart/

Best practices¶

• Start with the default of eight synthesis steps, then measure quality against latency. Tuning runs from 5–12 steps and speed from 0.7–2.0. — https://supertone-inc.github.io/supertonic-py/quickstart/
• Pre-download the roughly 400 MB model asset during CI or deployment setup so the first production request does not pay the download cost. — https://supertone-inc.github.io/supertonic-py/quickstart/
• Keep supertonic serve on loopback by default. Expose it only behind a reverse proxy. — https://supertone-inc.github.io/supertonic-py/quickstart/
• Separate the model licence and consent obligations from the MIT Python-source licence, especially when importing a custom voice profile. — https://github.com/supertone-inc/supertonic-py/blob/main/llms.txt
• Test screen-reader latency and language-specific number reading before accessibility deployment. A Turkish NVDA user reported delays and numeral-pronunciation failures after the v3 release. — https://github.com/supertone-inc/supertonic/issues/144

Superseded by this¶

• 2026-04-29 — Supertonic 3 replaces the five-language Supertonic 2 recommendation for new multilingual deployments. It supports 31 languages and keeps the v2-compatible public ONNX interface. Supertonic 2 remains in use for legacy integrations on its release branch.
• 2026-05-18 — Python embedding is no longer the only official integration path. SDK 1.3.1 added a local HTTP facade, while embedded use remains valid where simpler.

Still unknown¶

• We found no first-party Simplified-Chinese documentation in the checked Chinese search lane. Reviewed first-party material was English or Korean, which does not rule out such a page.
• PyPI lists both 1.3.0 and 1.3.1 on 2026-05-18. We found no release notes to treat 1.3.0 as a separate development event.

Sources¶

Agent brief¶

• Subject: project:supertonic, thread public-project-reference, 1 dated events 2026-05-18 → 2026-05-18.
• Practical note: Start with the linked public GitHub repository as of 2026-05-18. We have no evidence for version-specific adoption guidance.
• Confidence: high. Dated supersedes above are the authority for what is obsolete.