Tatiana Likhomanenko

Probably this could help https://github.com/facebookresearch/wav2letter/wiki/Inference-Run-Examples#interactive-streaming-asr-example. Or do you need to do this with w2l model, not inference model?

You can simply try to do your own main.cpp where you load the model and hang while communicating with some buffer where you get the data. I guess you can...

Please have a look at this tool https://github.com/facebookresearch/wav2letter/tree/v0.2/tools#voice-activity-detection-with-ctc--an-n-gram-language-model

As listed here https://github.com/flashlight/wav2letter/wiki/Building-Python-bindings you need to have Atlas or OpenBlas dependency installed.

My guess it is trying to use MKL version and you didn't install it. Could you set `export USE_MKL=OFF` and then reinstall w2l?

@danielkope, @Charmelink could you describe more details on your use cases and what is not working/which guide you need? Happy to help and explain.

Ok, I guess @vineelpratap knows more details on this. But yep, for streaming we used smaller model due to final restrictions on the performance for online inference.

Could you post which version of fl and w2l are you using?

We had problem with cpu backend before and it was not consistent with cuda (so cuda trained model didn't give the same output with cpu). This is fixed in current...

cc @vineelpratap @avidov @jacobkahn @xuqiantong Do we have converting scripts or any guides / hints how to do this?