Hi, I am trying to use the decoder after training on a sample dataset.
../Decoder --datadir data/ --lexicon trained_librispeech/enwiki-lexicon --lm trained_librispeech/enwiki-4gram.bin --test dev-clean/000000000.flac --am trained_librispeech/model.bin --tokens trained_librispeech/tokens.txt --show true --showletters true --logtostderr=1
The above is the command I am using to run the decoder, where "dev-clean/000000000.flac" is a sample audio file.
This results in "terminate called after throwing an instance of 'std::invalid_argument'"
Passing only the folder path, i.e: dev-clean/ (as mentioned in https://github.com/facebookresearch/wav2letter/blob/88f38d4de151d77de41cfef215ee18d6dbe4362e/tutorials/1-librispeech_clean/decode.cfg#L6) results in
I0704 11:34:16.600574 4404 W2lListFilesDataset.cpp:137] 0 files found.
I0704 11:34:16.600620 4404 Utils.cpp:102] Filtered 0/0 samples
I0704 11:34:16.600639 4404 W2lListFilesDataset.cpp:62] Total batches (i.e. iters): 0
I0704 11:34:16.600648 4404 Decode.cpp:138] [Serialization] Running forward pass ...
I0704 11:34:16.811210 4404 Decode.cpp:185] [Dataset] Number of samples per thread: 0
I went through the documentation, but couldn't figure out how to run the decoder.
Any suggestions are appreciated. Thank you.
I figured out that, passing a .lst file in the lua wav2letter's input format, decoder is working.
Most probably, the documentation is not up to date. Closing the issue.
Most helpful comment
I figured out that, passing a .lst file in the lua wav2letter's input format, decoder is working.
Most probably, the documentation is not up to date. Closing the issue.