whisperx failed transcription
hello,
i encountered following error with whisperx - any suggestions?
Jan
Performing alignment...
Traceback (most recent call last):
File "/usr/local/bin/whisperx", line 8, in
sys.exit(cli())
^^^^^
File "/usr/local/lib/python3.11/dist-packages/whisperx/transcribe.py", line 203, in cli
result = align(result["segments"], align_model, align_metadata, input_audio, device, interpolate_method=interpolate_method, return_char_alignments=return_char_alignments, print_progress=print_progress)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/dist-packages/whisperx/alignment.py", line 242, in align
trellis = get_trellis(emission, tokens, blank_id)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/dist-packages/whisperx/alignment.py", line 375, in get_trellis
trellis[t, 1:] + emission[t, blank_id],
~~~~~~~~^^^^^^^^^^^^^
IndexError: index 50 is out of bounds for dimension 1 with size 50
Hi!
Unfortunately, we do not have any experience with whisperx and using our W2V2 models with it. From the error traceback it seems like an off-by-one error. Did you try other models and do they work OK?
At the moment we sadly do not have much free resources available to invest into this problem, but please describe your setup, code, and problem in more detail, so that at some point we or other users can try to replicate and solve the situation.
hi,
with this model it works - at least for first try..
https://huggingface.co/classla/wav2vec2-xls-r-parlaspeech-hr