Alexander

There is a 30-40% time spent difference while inference on the same russian audio with -l ru and auto language detection(no -l )