You can deploy this model and start earning money today!

jinhwan000/
supertonic-3-mirror

Name: supertonic-3-mirror
Availability: InStock

Supertone/supertonic-3

quantized

openrail

Views

No views yet

Supertonic 3 — Preservation Mirror

This is an unofficial preservation mirror. It is not affiliated with, endorsed by, or maintained by Supertone Inc. All model weights are redistributed unmodified under the terms of the original BigScience Open RAIL-M License.

Original: Supertone/supertonic-3 · Copyright (c) Supertone Inc.

Why this mirror exists

On 2026-07-23 the upstream project announced that supertone-inc/supertonic would be archived, with no further development or official support for the open-source Supertonic models. This mirror exists so the released weights remain retrievable if the original repository becomes unavailable.

No claim of ownership is made. If the original repository remains available, please prefer it.

Provenance


Source repository	`https://huggingface.co/Supertone/supertonic-3`
Source revision	`3cadd1ee6394adea1bd021217a0e650ede09a323`
Revision date	2026-05-18
Mirrored on	2026-07-31
Copyright holder	Supertone Inc.

Integrity — SHA-256

Every model file in this repository is byte-identical to the source revision above.

File	SHA-256
`onnx/vector_estimator.onnx`	`883ac868ea0275ef0e991524dc64f16b3c0376efd7c320af6b53f5b780d7c61c`
`onnx/vocoder.onnx`	`085de76dd8e8d5836d6ca66826601f615939218f90e519f70ee8a36ed2a4c4ba`
`onnx/text_encoder.onnx`	`c7befd5ea8c3119769e8a6c1486c4edc6a3bc8365c67621c881bbb774b9902ff`
`onnx/duration_predictor.onnx`	`c3eb91414d5ff8a7a239b7fe9e34e7e2bf8a8140d8375ffb14718b1c639325db`
`onnx/tts.json`	`cf1a393b4f46c16d5af623228566d62658c7ef18bb524549c5ad70c583675121`
`onnx/unicode_indexer.json`	`9bf7346e43883a81f8645c81224f786d43c5b57f3641f6e7671a7d6c493cb24f`
`config.json`	`deeb55ddd865fe15730e323388c775ef7420dbe8d0f5fba452506203e3101b28`
`LICENSE`	`75d70017a6be9ba84e45af0d78f145a87fe372531d3e33076455dc520ce2cc4c`

Checksums for all files, including the ten voice_styles/*.json speaker styles, are in SHA256SUMS.txt. Verify with:

sha256sum -c SHA256SUMS.txt

Changes from the original

Per §4(c) of the License, changed files are disclosed here:

README.md — replaced with this mirror notice. The original model card is preserved verbatim as README_ORIGINAL.md.
SHA256SUMS.txt — added by this mirror.

No other file has been altered. All ONNX weights, tts.json, unicode_indexer.json, config.json, LICENSE, and every voice_styles/*.json are unmodified copies.

onnx/
  vector_estimator.onnx      244.6 MB   flow-matching core
  vocoder.onnx                96.7 MB
  text_encoder.onnx           34.7 MB
  duration_predictor.onnx      3.5 MB
  tts.json                               model configuration
  unicode_indexer.json                   unicode → token index table
voice_styles/                            F1–F5, M1–M5 (10 speaker styles)
audio_samples/                           12 reference / synthesized wav pairs
img/                                     hero image + benchmark figures
config.json  README_ORIGINAL.md  LICENSE  SHA256SUMS.txt

Inference code is not included here. See the original project supertone-inc/supertonic (MIT), which provides reference implementations in C++, C#, Python, Rust, Go, Java, Swift, Node.js, Flutter and Web.

License — BigScience Open RAIL-M

The weights are licensed under the BigScience Open RAIL-M License, dated August 18, 2022. The complete text is included in LICENSE. Read it before use or redistribution.

This is not an OSI-approved open-source license: it carries use-based restrictions.

If you use these weights, you must comply with the use restrictions in Attachment A of the License. Among others, the Model must not be used to violate applicable law; to exploit or harm minors; to generate or disseminate verifiably false information intended to harm; to generate personally identifiable information used to harm; to publish machine-generated content without disclosing that it is machine-generated; to defame or harass; to impersonate others without consent (including deepfakes); for fully automated decision-making affecting legal rights; to discriminate against individuals or groups; to exploit vulnerabilities of specific groups; to provide medical advice or interpret medical results; or for administration of justice, law enforcement, immigration or asylum processes. Attachment A of the LICENSE file is the authoritative and complete list.

If you redistribute these weights or derivatives, §4 of the License requires you to:

Include the use restrictions of Attachment A as an enforceable provision in any agreement governing use or distribution, and notify your recipients that they apply;
Give recipients a copy of the License;
Cause any modified files to carry prominent notices stating that you changed them;
Retain all copyright, patent, trademark and attribution notices.

Derivative works may be released under different terms, but those terms must include at minimum the same use-based restrictions as this License.

Per §6, the licensor claims no rights in the output you generate; you are accountable for that output and its subsequent uses.

Trademarks

The License grants no trademark rights (§8). "Supertone" and "Supertonic" are marks of Supertone Inc. and are used here solely to identify the origin of the mirrored files. This repository is not an official distribution and does not imply endorsement or successorship.

jinhwan000/supertonic-3-mirror