open-speech-corpora

Why we included this project

Anyone building speech models knows the hard part often comes before training: finding audio data you're actually allowed to use. This project collects openly licensed speech corpora into one indexed list, grouped by license so you can tell at a glance whether a dataset fits commercial use. Entries cover languages from Danish and Swedish to the large multilingual Common Voice set, with rough hour counts, speaker numbers, and direct download links for each. The issue tracker carries a long backlog of proposed additions, which gives a decent sense of what the community is still hunting for. It won't train a model for you, but it can save days of dataset scouting.