Support for speech to text models.

[JamesBowerXanda]

Feature request

Hi, it would be really useful if speech to text models could be supported by optimum, specifically to ONNX. I saw a repo that managed to do it and they claimed they used optimum to do it.

https://huggingface.co/Xenova/speecht5_tts

Is there a way to do this?

Motivation

I am finding it very difficult to convert any speech to text models to ONNX format and this would be very useful for both optimising serving them and also possibly running them with transformers.js.

Your contribution

I don't think I would be able to do this myself unfortunately.