That challenge seems to be more about speech command recognition (isolated words). They supply 1 second long recordings of 30 short words.

> There are only 12 possible labels for the Test set: yes, no, up, down, left, right, on, off, stop, go, silence, unknown.

This won’t work for large vocabulary continuous speech recognition, which is what you want if you want to transcribe podcasts, phone calls, or generally human-to-human, spoken interactions.