search

flashlight / wav2letter

Facebook AI Research's Automatic Speech Recognition Toolkit
https://github.com/facebookresearch/wav2letter/wiki
Other
6.37k stars 1.01k forks source link

How to convert W2L model to Pytorch? #956

Closed DongChanS closed 3 years ago

DongChanS commented 3 years ago

Question

There is a method that convert Pytorch (fairseq) model to W2L. (https://github.com/facebookresearch/wav2letter/tree/master/recipes/utilities/convlm_serializer)

Conversely, I want to convert W2L model parameter that pytorch supported. (ex, Transformer seq2seq in https://github.com/facebookresearch/wav2letter/tree/master/recipes/sota/2019)

Is there any convenient way?

tlikhomanenko commented 3 years ago

Nope, we don't provide such functionality, you need to convert by yourself (you can use example of convlm converting). You can find in other issue converting TDS model (somebody successfully tried that and published in the issue the code).

As there is no standard format (as some time ago ONNX was supposed to be) we don't support general format to convert models into other frameworks.