thewh1teagle/sherpa-rs

Multi-threading support

marchellodev opened this issue · 6 comments

When I'm running the vad example (without diarization, just vad) on a two-core linux server, only one core is fully utilized.

Is it possible to make use of all available cores?

Thank you for such an awesome library!

When using silero v5 and increasing num_threads to the number of cores, utilization is higher, but it is still about 20% on average, not 100

With v4 and v1 it's higher, more like 60% on average, but still it could be better

vad is super fast with a single thread. is there a reason to use multiple threads for vad?

@csukuangfj on a single thread (CPU) it takes way to much time to process a large audio (1.5 hours)

on a single thread (CPU) it takes way to much time to process a large audio (1.5 hours)

How long it takes to diarize 1.5 hours?
I remember when I worked on pyannote-rs it took me ~2 minutes for one hour.

@thewh1teagle around 5 minutes

Nevermind this issue, num_threads works great! Thanks for such an awesome library!