Batch inference

[thewh1teagle]

I use whisper.cpp in the project loud.cpp along with pyannote for diarization.
However I encounter the following issue:
most of the sentences are less than 30s and whisper process them as they were 30s and it makes the transcribe process much slower.
How can I batch inference segments in a way that I know which text belong to which batched segment?
Thanks