Open BeauregardTA opened 1 year ago
I think the importers here are for this purpose:
https://github.com/coqui-ai/STT/tree/main/bin
Shouldn't that phrase point to them? It is not only the audio format but the cvs files are also needed.
Well, the readme in that directory says that those are all for training models. Somebody with more knowledge of how the project works will have to tell me if they are also relevant to transcribing audio files.
Right. The whole workflow (train/test/inference) needs 16 kHz .wav files as input. Train and test (evaluate) also need .cvs files in a specific format, either prepared by the importers or custom scripts.
Maybe like this?
Currently, 🐸STT is only compatible with 16 kHz .WAV format. Datasets in other formats will need to be converted using the provided importers or custom scripts before they can be used in 🐸STT.
What do you and the maintainers think?
It also works with opus files in MLS I.e.
While this information is hinted at elsewhere in the documentation, is is only found in pages concerning training models. People who are looking to deploy pre-trained models are likely to miss this information and it should be explicitly stated in parts of the documentation that they are likely to look.