Is it possible to add audio context length parameter like in whisper.cpp

[fquirin]

I'm working mostly with very small files like 2-6s and what I noticed many times in Whisper is that transcription time is not linear. What I mean is that 3s and 6s files will be processed in almost the same time. I don't know the inner parts of Whisper well enough but I think this is related to the 30s window used to process chunks.

In whisper.cpp there is a parameter called audio_ctx that has been described here: https://github.com/ggerganov/whisper.cpp/issues/137

Do you think it might be possible to implement something similar in faster-whisper so that we can experiment with it and see if it reduces processing time for small files?