Supertonic 2 · int8 for the web
An int8-quantized copy of Supertonic 2 by Supertone, used as the optional "enhanced voice" of Lector MD, a Markdown reader that runs in the browser.
The model runs on the user's device with onnxruntime-web (WebAssembly, CPU, single thread). Nothing
is sent to a server.
What's here
| Path | What it is | Origin |
|---|---|---|
onnx/duration_predictor.onnx |
int8 | sherpa-onnx (duration_predictor.int8.onnx) |
onnx/text_encoder.onnx |
int8 | sherpa-onnx (text_encoder.int8.onnx) |
onnx/vector_estimator.onnx |
int8 | sherpa-onnx (vector_estimator.int8.onnx) |
onnx/vocoder.onnx |
int8 | sherpa-onnx (vocoder.int8.onnx) |
onnx/tts.json, onnx/unicode_indexer.json |
config and character table | Supertone/supertonic-2 |
voice_styles/F1.json … M5.json |
10 voice styles (5 female, 5 male) | Supertone/supertonic-2 |
The files are byte-for-byte copies of those sources (only the .onnx files were renamed). They are
mirrored here so the app doesn't depend on third-party repositories that could change or disappear.
About 96 MB in total, versus 263 MB for the original fp32 model. The inputs and outputs of each graph are the same as the original, so the official web example works unchanged.
License and use restrictions
The model is released by Supertone under the BigScience OpenRAIL-M license, included in
LICENSE. This copy, and anything built from it, is subject to the same terms, including
the use-based restrictions in Attachment A. Among others, the model must not be used:
- to impersonate others (for example, deepfakes) without their consent;
- to generate or spread verifiably false information meant to harm others;
- to exploit or harm minors;
- to defame, harass or discriminate against people;
- in any way that violates applicable laws or regulations.
Read the full list in LICENSE before using the model.
Credits
- Supertonic 2: Supertone Inc.
- int8 quantization: k2-fsa/sherpa-onnx
Model tree for mdreader-supertonic/supertonic-int8
Base model
Supertone/supertonic-2