Whisper.cpp is really nice software. I wanted to try making a plugin for Cinelerra with it, but I just haven’t gotten around to it yet. The main advantage is compiling whisper.cpp with CUDA on the user's PC. Even an old RTX 2060 can transcribe hour-long recordings very quickly. Best regards, Andrey чт, 3 сент. 2026 г. в 14:28, Andrew Randrianasulu via Cin < cin@lists.cinelerra-gg.org>:
I was trying to find any use for those large language models and found whole site with autotranscribed WWDC videos.
I looked into whisper.cpp and after some bugreporting it installs and works even on 20 years old processor!
https://files.mastodon.online/media_attachments/files/117/206/767/104/645/63...
I used base model:
wget https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin\?download=true
and "jfk.mp3" sample (real classic!):
wget https://github.com/ggml-org/whisper.cpp/raw/refs/heads/master/samples/jfk.mp...
I think in my case without any acceleration and just 2 2GHZ cores it processed 11 seconds in 126 or so seconds,so def. not realtime, but
a) there are turbo variants of model
b) usually people have more powerful machines with gpu, so may be it can be used for auto-transcribing some long lecture or somesuch.
This speech-to-text was promised LONG ago and I guess only now, after serious sacrifices in terms of energy and other damages by Big Bro Tech, actually becoming usable ... :/
But I figured if I used OCR for all those 25 years I can use this STT engine too, at least for testing.
_______________________________________________ Cin mailing list -- cin@lists.cinelerra-gg.org To unsubscribe send an email to cin-leave@lists.cinelerra-gg.org