Long Form Audio
#10
by
deathknight0
- opened
Thanks for the great model! Just curious, the model's accuracy is great but for some reason it does not punctuate long sentences (eg comma, full stop except at the end of the transcription). I looked at the model's training dataset and looks like there are some long form audio in it (but not alot - I did not scrutize it deeply admittedly). Just wondering if the model was designed to handle long form audio..?
Hi, thank you for finding our model useful. Indeed, our model doesn't currently produce punctuation and casing because some training corpora didn't have that information. Not an ideal solution but you could try to restore casing and punctuation with an additional LLM pass that doesn't change the word sequence.
gsaon
changed discussion status to
closed