Is there a way to download (through API or something) current clips? Besides from the Dataset page.
No. What are you trying to do? Get the “new recordings” to build a Delta?
PS: Current API is not public and mostly designed to be used by the server code itself…
No, no such ambitions . We want to make MVPs; two versions to compare the quality. But mostly to motivate volunteers to do more. Now we have 100+ hours, but the latest dataset is 49 hours.
Sorry for my ignorance, what is a MVP? Minimum Viable Product? Of what kind?
Minimum Viable Product. A kind of raw program that does the minimum work.
So, you create a model with v12 data and then another with v13 data and look at the CER/WER (or whatever your modelling tool supports) results first. Then you apply them in the real-world to be tested…
Why would you need API access for this?
BTW, I did the same with this project last year (somewhat archived for now). If you want, you can easily add your language there and test your coqui-STT model…