Transcribe one hour in less than 10 minutes!

[thewh1teagle]

I found a way to transcribe in speed similar to paid APIs, locally on your computers (MacOS)

1. Download ggml-medium-encoder.mlmodelc.zip
2. Open models folder and unzip it there (filename should be ggml-medium-encoder.mlmodelc)
3. Transcribe some file (first time takes a while but only the first time)
4. Transcribe in the speed of the light ⚡️

Turn sound on

tauri.mov

[Willxiam]

that is awesome. I tried on a windows machine and I cannot tell if it makes it much faster. All I did is put the unzipped file in the model directory.

[Willxiam]

I tried it on the video demo above, which is only 38 seconds and it did not increase the speed. So I do not think this works on windows 10, at least not with the way I did it. I did not change any settings in vibe, I just unzipped the file into the folder

[thewh1teagle]

I wrote that it works on MacOS only on top of this page

[Willxiam]

Yeah I am sorry. I guess I was posting in case someone else was curious and I was testing because I was hopeful, that you specified MacOS because that is what you had tested it on, not because it did not work on Win. Hopeful cause I would love to be able to transcribe faster.

[thewh1teagle]

@Willxiam
I added GPU optimization for Windows in version 1.0.0, it should work x1.5 faster on computers with only integrated gpu,
and on computers with nvidia card should be even much faster.