undefined | Better HN

0 pointsStevenWaterman3y ago0 comments

Their models range from 70mb to 3gb. The largest model is smaller than the optimised stable diffusion. Not sure what the inference speed is like, haven't tried it myself yet.

0 comments

IceWreck3y ago

I just tested it myself. Its fast enough on colab, couple of seconds but not sure if its fast enough to transcribe realtime audio yet.

lynguist3y ago

"small" runs in realtime on Macbook Air M1 CPU.

MacsHeadroom3y ago

Colab is using one of the larger models. Tiny probably runs in realtime on a single core of an RPi.

j / k navigate · click thread line to collapse