I'm not far enough into my project to know for sure, but I think that, for example, using tract to do speech recognition can be done in sub-realtime (ie, the in

I'm not far enough into my project to know for sure, but I think that, for example, using tract to do speech recognition can be done in sub-realtime (ie, the inference takes shorter than the audio duration) on a Macbook M1 Pro. Hopefully on less powerful devices too, but I haven't tested that yet.