For the last year I was researching and developing an ML model that understands speech with disorders, like speech of people after strokes or with dysarthria. I think it's a right time to launch early, so if you know a person who struggles with voice communication please let them know
App works in the browser and should work on any device
From a technical perspective, it's a PEFT (LoRA) fine-tuned version of distilled Whisper on all available data for this task, with some data augmentations, trained for about a day on a single RTX 5090
This is very early stage so things will be often broken and the model will be often updated, but if you are not afraid to experiment, I invite anyone with speech problems to try
I'm happy to answer all questions
Comments URL: https://news.ycombinator.com/item?id=44249789
Points: 1
# Comments: 0