AI and music

It's been able to compose and arrange for decades, but even simple transcription is hard. There's no way for it to quantify gestures except via pattern matching. Manny's experimenting with Sona, which is producing some very credible demos, but it does alter the contours and the rhythms to fit the predictive scheme. Fourier transforms might reveal something of the underlying harmonic series of a waiata, but I doubt that they'd be much use for transcribing waiata. It happens in space and time, so getting some essentialised figure isn't useful anyway.