>>40231242>>40232441If I had to guess, he created a custom controllable talknet singing model using her voice. Now whenever he wants her to sing a new song:
he'll take a target song
>use a program to separate the vocal and instrumental parts of the song >use the custom controllable talknet model to convert the vocal part of the song he extracted into a version using her voice >use an audio program to piece the song together again using her custom vocals and the instrumental part of the song 