Pipe the output of ffmpeg to cat before reading it.

[csukuangfj]

See https://github.com/lhotse-speech/lhotse/issues/666#issuecomment-1100574606

[danpovey]

After looking at the size of reads, it seems that this does not resolve the problem. I suspect it was not really faster, e.g. we were comparing across machines with different properties. We should double-check though; if it was faster, it is worth investigating why.

[pzelasko]

Let me know once you think it's good to merge it.

[pzelasko]

Did you reach any conclusions? I can just merge it as it won't break anything, but it would be good to know if it really helps.

[danpovey]

Personally I don't believe this can be helping, even though it appeared to. From strace output, it doesn't seem to affect the chunk sizes it's read in; and I don't believe cat actually does any buffering. But if the effect can be replicated I suppose we'd have to find an explanation.