Need help - model producing partial transcriptions.
#3
by
IAmNotAnanth
- opened
while testing, the model produced just a few words, or just one sentence for any duration of audio input.
Also for context, the model was giving Array Out of Bounds exceptions as it was trying to access suppress tokens.
As a workaround, I used the suppress tokens from the native model. After multiple attempts, the model failed to produce a complete transcription.
I was able to use the same code and produce a complete transcription with the Hindi model weights provided by you.
- is there any fix for this issue, and how reliable it the presented WER of 10.800
- Is the model capable of producing time stamps?
- Is it possible to run translation task on the model or is it purely geared at transcriptions only?