supertonic-3-spectalo

Supertonic 3 — mirror for on-device apps

A fixed set of Supertonic 3 files, kept here so that Calida Lab's apps (SpectaBooks, SpectaLing) can download a known voice to the device. The apps pin a commit of this repository and check every file against a SHA-256 before using it; nothing here is fetched from a moving branch.

Supertonic 3 is made by Supertone Inc. This repository only redistributes it. Calida Lab has not modified any file.

Files and where they come from

Every file is byte-identical to the file named as its source (compared by SHA-256).

Modified files

vector_estimator.int8.onnx and vocoder.int8.onnx are modified versions of Supertone's onnx/vector_estimator.onnx and onnx/vocoder.onnx: they were quantized to int8 by the k2-fsa project (sherpa-onnx), which is what makes the whole voice about 145 MB. They are redistributed here as k2-fsa published them.

Licence

The model is licensed under the BigScience Open RAIL-M License — see LICENSE, a verbatim copy of the licence in Supertone/supertonic-3. It applies to every file in this repository and to anything derived from them.

The licence includes use-based restrictions (Attachment A) that bind everyone who uses the model or its output. Among them, you may not use it to impersonate others without their consent (for example, deepfakes), to generate or spread verifiably false information in order to harm others, to generate content without clearly disclosing that it is machine-generated, or in any way that violates applicable law. Read Attachment A in LICENSE for the full list; the licence text, not this summary, governs.

Attribution

Supertonic 3 — © Supertone Inc. — https://huggingface.co/Supertone/supertonic-3 · https://github.com/supertone-inc/supertonic int8 quantization of the vector estimator and vocoder — k2-fsa / sherpa-onnx.