exactly the way that he used to speak. text—to—speech technology has been around for a while. we all remember microsoft sam, right? sam speaks monotonously: h 3 voice sounded a tad monotonous. peter provided far more than the average voice sample recording amount — 15 hours' worth of material. in order to make a voice, you take that audio and you use a neural net system to learn how the sounds in the voice relate to the words that are being spoken. once you have that model built, you can put a new set of words in and it will produce a new output with that voice. to better convey sentiment for specific scenarios, peter captured himself speaking in four different styles. tone of voice is very important. tone of voice is very important! tone of voice is very important. he also tirelessly recorded 3,000 stock phrases, which he can quickly