4 ms·
I've definitely thought about how one goes about coding an intelligent agent that can identify what you're playing and the make accurate guesses about where in
by darylteo 2y ago
I've definitely thought about how one goes about coding an intelligent agent that can identify what you're playing and the make accurate guesses about where in the piece you're playing.
The player could make mistakes as well, which means any intelligent algorithm would need to apply some degree of probabilistic calculations to find the most likely point.
Very often, when practising, you'd constantly repeat passages multiple times. And certain passages may also be exactly the same in different parts of the piece as well.
To me it sounds like a incredibly difficult problem to solve. Not sure if the recent AI advances change anything.
- smokel 2y agoThere are multiple parts to this problem, but it seems quite doable. Shazam can do generic look-up of music since around 2003, see for example this page on abracadabra [1]. I'd guess that if you limit things to a piano, one could relatively easily find the notes being played with some Fourier magic, and then solve a Hidden Markov Model to find the most plausible position in the music. Using a MIDI interface makes it even simpler. [1] https://www.cameronmacleod.com/blog/how-does-shazam-work https://www.cameronmacleod.com/blog/how-does-shazam-work
- darylteo 2y agoOne day I hope to be so mathematically inclined to understand the math behind that.