Using pre-trained models with simple_streaming_asr_example

[pletessier]

I have successfully followed the tutorial to run simple_streaming_asr_example with the models dowloaded from http://dl.fbaipublicfiles.com/wav2letter/inference/examples/model/

I am now trying to use the pre-trained models presented in https://github.com/facebookresearch/wav2letter/blob/master/recipes/mls/README.md.

Is that compatible ?

I have replaced :

• tokens.txt with https://dl.fbaipublicfiles.com/wav2letter/mls/french/tokens.txt
• acoustic_model.bin with https://dl.fbaipublicfiles.com/wav2letter/mls/french/am.bin
• lexicon.txt with https://dl.fbaipublicfiles.com/wav2letter/mls/french/joint_lexicon.txt
• language_model.bin with 5-gram_lm.arpa from https://dl.fbaipublicfiles.com/mls/mls_lm_french.tar.gz (converted with build_binary)

I have an error when running simple_streaming_asr_example :

Started acoustic model file loading ... 
terminate called after throwing an instance of 'cereal::Exception'
  what():  Error while trying to deserialize a polymorphic pointer. Could not find type id 3
Aborted (core dumped)

acoustic_model.bin and https://dl.fbaipublicfiles.com/wav2letter/mls/french/am.bin are obviously not serialized in the same format.

[vineelpratap]

Hi, they are not compatible. It only supports models trianed with this recipe https://github.com/facebookresearch/wav2letter/tree/master/recipes/streaming_convnets/librispeech as of now...?

You can however, follow the commands in https://github.com/facebookresearch/wav2letter/blob/master/recipes/mls/README.md#decoding to do inference (but it won't be streaming) ....