Hey folks.
Thanks for all the effort you've put into this problem so far. After discovering whisper's poor WER rate for Indic languages, a deep dive brought me here.
Now ideally I could load this model into whisper libraries that I'm already using, but as I understand it they have different underlying architecture's and that's not possible.
So instead, I'm looking for an update to https://github.com/AI4Bharat/IndicWav2Vec?tab=readme-ov-file#huggingface-inference
I would like to use IndicWav2Vec Large to minimize the need to have separate models for each language - but I only see the fairseq model and I need to use it in my code rather than in a terminal.
How does one actually use this?
I see https://github.com/AI4Bharat/IndicWav2Vec?tab=readme-ov-file#deployment
but models_dict.json seems to just be a mapping to the smaller models.
Hey folks. Thanks for all the effort you've put into this problem so far. After discovering whisper's poor WER rate for Indic languages, a deep dive brought me here. Now ideally I could load this model into whisper libraries that I'm already using, but as I understand it they have different underlying architecture's and that's not possible. So instead, I'm looking for an update to https://github.com/AI4Bharat/IndicWav2Vec?tab=readme-ov-file#huggingface-inference I would like to use IndicWav2Vec Large to minimize the need to have separate models for each language - but I only see the fairseq model and I need to use it in my code rather than in a terminal. How does one actually use this? I see https://github.com/AI4Bharat/IndicWav2Vec?tab=readme-ov-file#deployment but models_dict.json seems to just be a mapping to the smaller models.