caustik

I agree

I was able to optimize short-term and momentary full-file analysis by chunking the audio across the number of available threads and then adding some "preroll" aligned to the block boundary...