3 ms·

okay, but.. any path to connecting this to phone mics so that it can function anywhere? whats the state of the art in tone and chord recognition models?

• jasinjames 2y ago
  No need for any ML models here, you can "just" do an FFT on a conventional signal and pick out the peaks
  • swyx 2y ago
    (i mean yes i agree but) so ... why do i feel like ive never seen anyone do this before?
    • udit99 2y ago
      If you have a guitar handy, you can try out the live mode on https://www.fretboardfly.com https://www.fretboardfly.com . It works and people use it: Using your acoustic guitar to play the fretboard learning game via the mic.
    • criddell 2y ago
      I believe this is how Yousician works.
  • codyd51 2y ago
    The waveform produced by sounding a note on most physical instruments will often not exhibit a peak, or will not exhibit the strongest peak, on the note being sounded. Rather, most instruments will instead produce harmonic overtones and our brains fill in the gap of the pitch that’s intended to be sounded. You can still absolutely deduce the fundamental with great accuracy via an FFT, but the approach is a bit more involved. The relevant research area here is called ‘fundamental frequency estimation’. For an example of this, you can see this app I built that lets you give keyboard and mouse inputs via playing notes on a bass guitar, which are recognized over the microphone: https://github.com/codyd51/offkeyboard https://github.com/codyd51/offkeyboard