Because F Yousician

Discussion in 'TZT GameDev' started by AgelessDrifter, Dec 4, 2017.

  1. AgelessDrifter

    AgelessDrifter TZT Neckbeard Lord

    Post Count:
    42,891


    I'm writing my own app to practice sight reading sheet music

    Still needs some tweaking and a UI but it's approaching passable.

    Would like to get it to take in sheet music images and display the correct note durations/ play at the correct speed so that if you always hit on the red line you'll be playing in the correct time, but that's a ways off. Not sure it'll play well with chords the way the audio recognition is written, and including them would take some adjustments in the logic

    Anyway it's already in a sufficient state for me to use it to practice quick note recognition/muscle memory in all the possible scales. Should work for any instrument, too, as long as it's loud enough and in tune.
     
  2. Utumno

    Utumno Administrator Staff Member

    Post Count:
    38,107
    Ageless Rocksician
     
  3. AgelessDrifter

    AgelessDrifter TZT Neckbeard Lord

    Post Count:
    42,891
    Turns out chord recognition is a huge open problem in computer hearing :hmm:

    Can't even get this thing to recognize the lower notes on the keyboard individually, I discovered--not sure what's up there. If the sampling resolution of the discrete fourier transform is high enough to recognize the high frequencies I'd think it should spot the lowest ones pretty handily, but I guess that's not how it works
     
  4. Agrul

    Agrul TZT Neckbeard Lord

    Post Count:
    44,411
  5. Yogr

    Yogr Meatball Sub Pro

    Post Count:
    1,162
  6. AgelessDrifter

    AgelessDrifter TZT Neckbeard Lord

    Post Count:
    42,891
    Hmm yeah that is pretty bad

    Still haven't worked out what's going on with the high and low frequencies--I wouldn't think harmonics would be much of an issue on an electric keyboard (relative to, say, an electric guitar) but I suppose it isn't exactly midi audio, either

    A friend who's into recording stuff tells me he bets it's just that the mic is too shitty and is gonna bring by some better equipment to help test that hypothesis

    Still, I feel like there's gotta be some mathemagic I can work on this so it'll work ok on typical built-in microphones--compressed sensing or convolution or some shot--without having to sacrifice sufficiently rapid response time