680,000 hours of multilingual audio is what OpenAI trained Whisper on: 438,000 hours of English speech, 126,000 of non-English speech across 68 languages, and roughly 125,000 hours of translation data. Filecoin network operates at exabyte scale, where this is routine.

680,000 hours of multilingual audio is what OpenAI trained Whisper on: 438,000 hours of English speech, 126,000 of non-English speech across 68 languages, and roughly 125,000 hours of translation data. Filecoin network operates at exabyte scale, where this is routine.

Source:https://x.com/Filecoin/status/2087625693790077217