I tested it in danish and it works quite well in total. When there is silence though it seems to create loads of identical markers (40+) with the same dialogue attached… I can delete those, so its okay, but the main problem is that the created marker tracks makes navigating my session very slow. Perhaps a graphic issue? I tried to deactivate marker lines across all tracks but with no performance gain.
Well thank you for introducing me to Viegas! I watched some of his recitations, and they’re indee spectacular! Of course here in Brazil we study Camões, Fernando Pessoa, but I have to be honest that it all makes much more sense coming from a portuguese person.
In my case, although the material was full of words from indigenous languages - things I believe the models had very little or none at all to work with - it still seemed to make phonetical approximations that were pretty close. But in my case it was a narration, intended to be as clear as possible, so that must make things a bit easier on the detection or something.
I can imagine that there’s more brazilian portuguese around to train the models on, but it still seems strange that it didn’t catch anything at all, right? I mean, as thick of an accent Mário may have, still many words are pretty clear to my brazilian ears… So I’m at a loss at what goes on under the hood here.
1 Like