README.md

---
license: openrail
license_name: bigscience-openrail-m
license_link: LICENSE
pipeline_tag: text-to-speech
tags: [tts, onnx, supertonic, mirror]
base_model: Supertone/supertonic-3
---

# Supertonic 3 (mirror) — for the arraypress speech tools

An **unmodified** mirror of [Supertone/supertonic-3](https://huggingface.co/Supertone/supertonic-3)
at commit `3cadd1ee6394adea1bd021217a0e650ede09a323`: the six ONNX/JSON files of the
pipeline (duration predictor, text encoder, vector estimator, vocoder, tokenizer config,
unicode indexer) and the ten voice styles (F1–F5, M1–M5), ~410 MB. It exists so the fleet's
`swift-tts` (and the `speak` / `redub` tools on it) download from a repository pinned and
controlled by the same people as the tools, rather than following another repo's `main`.

Every file matches the hash Supertone publishes (SHA-256 for the large files, git blob
SHA-1 for the small ones). For the model itself — what it is, quality, languages — see
[Supertone's model card](https://huggingface.co/Supertone/supertonic-3).

## Licence and use restrictions

**BigScience OpenRAIL-M** (`LICENSE`). Using these files means accepting it, including the
**use-based restrictions in Attachment A** — among them: no use that violates law; no
exploiting or harming minors; no generating or disseminating verifiably false information
to harm others; no personally identifiable information to harm individuals; no
defamation or harassment; no fully automated decisions that adversely affect legal rights;
no discrimination; no medical advice or results interpretation; no use for administration
of justice or law enforcement in ways that adversely affect rights. Read Attachment A for
the full list. Anyone you share the model or a derivative with must be bound by the same
restrictions. Attribution: see `NOTICE`.