41–50 of 93 posts
Awesome work. Honest question, do you think with this marked improvement it could be worth making a wrapped library for this c/c++ version in python - for example like numpy?
CPU inference is not cost effective for my use-case
Is there any improvement to GPU inference at all? CPU inference is not cost effective for my use-case
There are tons of open source STT models, what makes whisper so valuable? I especially don't get it on mobile, where the native STT built into the OS is now real-time and includes punctuation (at least for iOS). I love the open-source approach to the model, but it didn't strike me as particularly better than other open-source or built-in models.
In my limited testing, I've found Whisper to be much better (accuracy-wise) than other STT models.
I vouch for this. Pretty solid and keeps improving. The OP is in the class of Magic Wizards of programming like Fabrice Bellard! There are frequent updates and performance improvements. There is also a small community of active users around this. All most all feedbacks get implemented and the OP is very responsive. The OP made it possible to do state of the art voice recognition without the PyTorch baggage and in C/C…
A single binary that does this all w/o the python stack is just incredible!
edit: Got it going in 1 min!
I grabbed the prebuilt artifacts (windows)
- https://github.com/ggerganov/whisper.cpp/actions/runs/363552...
Then downloaded ggml-base.bin (148mb) and put it in models/ggml-base.en.bin
- https://huggingface.co/datasets/ggerganov/whisper.cpp/blob/m...
Ran it and everything worked! Amazing. Note that only the large (3GB) whisper-v2 is available at the moment, but haven't seen any errors yet from the older small ones. Wild.
When I tried the medium model here it seemed to make more mistakes than the GPU version. Any idea why that would be?