Transcription speed is equally affected by encoding and decoding speed

[vadi2]

Just sharing my insights in case someone else runs into this:

#89 seems to suggest that encoding speed is king, but I suspect I've read it wrong: the reason we're only benchmarking encoding is because that is the only component that can be accelerated, not because that's the only part of the pipeline that matters.

I'm saving this because for decent-sized real world audio (say 15min), having both a strong CPU and encoding acceleration is important: not just encoding acceleration alone. I've found that what my ryzen 5+rtx 4080 was able to gain in encoding speed, it lost in decoding speed compared to the m1 pro and the total transcription time was just about the same.

Would be glad to hear if I'm on the right track with this.