notes.txt

voice timbre + volume = harmonics countour
Base: "impulse train" in the freq domain. Strict harmos, mag = 1. 

linear interp on log or not? 
    can't hear a difference. Using no-log

Try: 
    don't implement vibrato
    Sing with vibrato
    see what you get

hybrid synth
    low freq: harmoSynth
    high freq: ifftSynth
    
    to test:
        the output multiplier. Synth the same pitch with ifft and harmo: are they at the same level? 

Why does the sound sound loose when in triad? 
    test adding sounds together in Jupyter. 
    maybe related to irrational harmonics not interfering? 
    OMG instead of sort with freq i was sorting with mag

Spectral leakage: just sum the neiboring freq bins? 
    no. Sum is smaller when the freq is in the mid of a bin
    Sum of squares: look likes it! 
    So let's dump SFT
    results:
        yup it works perfectly

"with a periodic signal for voiced sounds and white noise for unvoiced sounds. "
    - Ning Hu
    YinYin! Take out the harmonics. Use the rest to synth the envloped noise. 
    test this in jupyter notebook with a pre-recorded "mouthhh fffrequencccccy"
    x sh f s th ts ch ha wu
    WHAT A SUCCESS

Key velocity! 
    Dannenberg, R. B. (2006, November). The interpretation of MIDI velocity. In ICMC.

TODO
    delay between tracks? to make it rich and less robotic

implement vibrato?
    - slow approach to a new note? 
    - perlin noise to add FM variations? 

Gus: 
    被动变色 (变色龙) 主动变色 (取色笔)

related work
    Aeolis: A Virtual Instrument Producing Pitched Tones With Soundscape Timbres

Mystery:
    When C3 is sang, how do people hear C4?