Apple Mac Mini M2 16GB - Transcription performance suggestions?

[heathdutton]

Hi, been watching & loving this plugin development.

However, I've been struggling to find the best possible settings for this on an M2 Mac 16GB.
This machine has one job, take an NDI feed, add captions (english and spanish using localvocal 0.3.9) and then feed back out via NDI for consumption elsewhere.

When I go with a small model, it's very responsive but terrible in quality (hallucinations, repetitions, wrong words) when I compare to Google transcriptions.

When I reach for larger models the transcription text improves but falls behind, skipping entire sentences, etc. The strange thing is that when using a larger model, the CPU cores (efficiency and performance) never go over ~50% avg, so I'm uncertain where the bottleneck is occurring precisely. The CPU gage for OBS never seems to go higher than 20%.

Should I just give up on this, and use a PC with a proper GPU or is there a suggestion/idea I am missing? I'm trying to use "Whisper Distil Medium English Q5 0 (Mukowaty)" or better, fyi.