search

flashlight / wav2letter

Facebook AI Research's Automatic Speech Recognition Toolkit
https://github.com/facebookresearch/wav2letter/wiki
Other
6.35k stars 1.02k forks source link

Use seq2seq librispeech models #975

Closed giannisp-verneek closed 3 years ago

giannisp-verneek commented 3 years ago

I am trying out the wav2letter++ example that you have using the docker image from the inf build branch. How can I use the models that are here: https://github.com/flashlight/wav2letter/tree/master/recipes/seq2seq_tds/librispeech with that version? I'm trying to load them and I'm getting the following message when the language model tries to load:

Loading the LM will be faster if you build a binary file.
Reading /root/Git/Projects/model/lm_librispeech_convlm_14B.bin
----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100
terminate called after throwing an instance of 'lm::FormatLoadException'
  what():  /tmp/kenlm/lm/read_arpa.cc:65 in void lm::ReadARPACounts(util::FilePiece&, std::vector<long unsigned int>&) threw FormatLoadException.
first non-empty line was "
tlikhomanenko commented 3 years ago

You are trying to use convlm model not the kenlm, so you need to specify flag on language model to be convlm or please put lm path as some kenlm model (bin or arpa).