Spaces:

Duplicated from  whisper-event/whisper-demo

sanchit-gandhi

/

whisper-large-v2

like 170

Running on T4

I tried to implement the same gradio app in Colab pro but I got different transcripts for the same audio clips. How can I reproduce your environment?

Can you list the version of transformers used and any generation parameters?

@sanchit-gandhi

I think I found the issue. Using return_timestamps=True gave me the same output as this space. So I guess Whisper is affected by timestamping.