Charalampos

/

whisper-large-el

like 0

Thanks in advance for fine-tuning for Greek and apologies in case my question is too basic! I am new here. :)

I am using the model via transformers AutoModel and AutoProcessor and I'd like to transcribe a long audio (~11mins).
To my disappointment, I only get a transcription of up to the very first seconds of the audio file (9-10s). I have experimented with the options max_new_tokens, and no_speech_threshold and also removed any long silence from the audio, without resolving the issue.
I previously used the multilingual OpenAI model via the whisper cli command and the entire audio was transcribed, so I am wondering if you have any idea what could be causing the generation/decoding to stop early.

Any advice is very welcome!