I'm working on a machine learning project, where you take voice recordings and teach a model, to speak as that person. That project have a perfect sound dataset and not what I'm asking about here. (this is a good eksemple of what I'm working on)
But after this project, I want to try the same thing with some podcasters that I been listening to. I have 274 mp3 podcasts (829 hours), that all contain 3 to 4 people (maybe a guest now and then). The recordings/podcasts are really clean in audio, but a lot of times they talk over each other (which can't be used in the dataset).
I want to be able to use some kind of software, that can detect who is talking (by their voice) and when. That way I can create a dataset, that have clean audio of only that person's voice. I could do this manually, if we didn't talk about 829 hours of recordings.
I don't need the software to detect all of them at ones (but it would be nice), but if the software can detect one voice out of all of them, then I can just repeat those step for each person.