Up until now, I’d always assumed it was basically just transcribing what I said, feeding the text into the model, and reading the response back. So out of curiosity, I asked if it could actually tell how I was speaking, not just what I was saying. So I just started throwing random tests at it. I whispered a sentence. It immediately picked up that I was whispering and even described the kind of tone I was going for. Then I switched to a dramatic loud voice, and it caught that too. Next,...