Add support for Supertonic TTS

[xenova]

Links:

• Original model: https://huggingface.co/Supertone/supertonic
• Original demo: https://huggingface.co/spaces/Supertone/supertonic#interactive-demo
• GitHub repo: https://github.com/supertone-inc/supertonic

---

Example usage:

import { pipeline } from '@huggingface/transformers';

const model_id = 'onnx-community/Supertonic-TTS-ONNX';
const tts = await pipeline('text-to-speech', model_id, { dtype: 'fp32' });

const input_text = 'This is really cool!';
const audio = await tts(input_text, {
    speaker_embeddings: 'https://huggingface.co/onnx-community/Supertonic-TTS-ONNX/resolve/main/voices/F1.bin',
});
await audio.save('output.wav');

output.mp4

---

TODOs:

• Add to list of supported models
• Create web demo
• Add model quantizations
• Use default speaker embeddings?

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[nico-martin]

Looks good to me!
I think its great that users can set the num_inference_steps, since it has a huge impact on the quality :).
As sugested in the comments I would make the speaker_embeddings optional and set a default. For now its not easy to create a custom voice so people will mosk likely have to use a default voice anyways. So I would add a comment on how they can switch between one of the four default voices but they dont have to set specify a speaker_embeddings.

[xenova]

For now its not easy to create a custom voice so people will mosk likely have to use a default voice anyways. So I would add a comment on how they can switch between one of the four default voices but they dont have to set specify a speaker_embeddings.

Sounds good! 👍