All Voice-to-text pipelines for linux sucked, so I made my own. I also patched upstream libraries like ONNX to actually work with latest models (was a PITA to make parakeet work!)
It's now the fastest and most accurate linux transcriber.
(fully FOSS of course)
The learning part seems nice. But I am confused from the small description I read on the website.
What's my involvement in its learning process? Do I have to provide fixes for words so that it knows something was mistranslated first?
No, you don't! It's fully automatic training based on your local voice clips. Nightly or Weekly, when your PC is idle, it uses a big fat model to transcribe your clips, learn, and train the faster smaller everyday model.
You CAN do manual training and corrections; it's helpful, and the hooks are there, but totally unnecessary. I don't do it, it's good enough as is