HN Explorer

Ask HN: Which speech to text model would you recommend?

1 points by spacetimeuser5 ↗ HN
I may need to perform a bit of speech-to-text (English at least, but in perspective - multilingual also) from video or audio files. Which speech-to-text model/API would you recommend, which sort of performs the best and can also do noise etc reduction?

4 comments

[ 3.4 ms ] story [ 24.0 ms ] thread
smoldesu
Whisper, 100%. It's small, fast and does a really good job with most of the recordings I can feed it. IIRC, there are both English and mixed-language models to choose from as well.
spacetimeuser5
This one https://huggingface.co/openai/whisper-large-v3 ? Do I need (and where from) to use CUDA from some google.colab or 4-core AMD Ryzen 3 4300U CPU on a laptop will handle it for an initial test program?
smoldesu
That one should work. The model itself is extremely fast without any acceleration, in my experience. The AMD CPU should perform just fine. Good luck!