Instructions to use amcic/supertonic-3-fp16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Supertonic
How to use amcic/supertonic-3-fp16 with Supertonic:
from supertonic import TTS tts = TTS(auto_download=True) style = tts.get_voice_style(voice_name="M1") text = "The train delay was announced at 4:45 PM on Wed, Apr 3, 2024 due to track maintenance." wav, duration = tts.synthesize(text, voice_style=style) tts.save_audio(wav, "output.wav")
- Notebooks
- Google Colab
- Kaggle
Supertonic-3 fp16 ONNX (availability mirror)
This repository redistributes existing model weights unchanged. It exists so that a published mobile application has a distribution endpoint whose availability we control. No model was trained, fine-tuned, or modified here.
Provenance
| Original model | Supertone/supertonic-3 — Copyright (c) 2026 Supertone Inc. |
| fp16 ONNX conversion | Kyumdroid/supertonic-3-quant |
| This repository | Byte-identical copy of the above conversion. No files were altered. |
The fp16 conversion was performed by Kyumdroid using
onnxruntime.transformers.float16.convert_float_to_float16 with keep_io_types=True
(fp32 I/O for SDK compatibility) and op_block_list=['Cast']. See the
upstream conversion notes for detail.
Every file in this repository has been verified by SHA-256 against the upstream conversion repository. All 16 files match.
Layout
fp16/onnx/
text_encoder.onnx
duration_predictor.onnx
vector_estimator.onnx
vocoder.onnx
tts.json
unicode_indexer.json
voice_styles/
{F1,F2,F3,F4,F5,M1,M2,M3,M4,M5}.json
voice_styles/ is unchanged from upstream.
License
BigScience Open RAIL-M, inherited from Supertone/supertonic-3. The full license text is included in this repository: LICENSE.
Copyright (c) 2026 Supertone Inc.
Use restrictions (notice to subsequent users)
The Open RAIL-M license carries use-based restrictions set out in Attachment A of the LICENSE. These restrictions apply to you, and you must pass them on to anyone you distribute the model or its derivatives to. In summary, the model may not be used:
- to violate any applicable law, or to exploit or harm minors;
- to generate or disseminate verifiably false information with intent to harm;
- to generate or disseminate personally identifiable information without consent;
- to defame, disparage, harass, or impersonate individuals;
- for fully automated decision-making that adversely affects legal rights;
- to discriminate against individuals or groups based on protected characteristics;
- to provide medical advice or interpret medical results;
- for law enforcement, immigration, or asylum processes covered by the license.
Read Attachment A of the LICENSE for the authoritative and complete list. The summary above is not a substitute for it.
Model tree for amcic/supertonic-3-fp16
Base model
Supertone/supertonic-3