Source-filter model: a glottal source (pulse train with -12 dB/octave tilt, or noise for whisper) drives a cascade of three digital resonators (Klatt):
y[n] = A x[n] + B y[n-1] + C y[n-2]
C = -exp(-2 pi BW / fs), B = 2 exp(-pi BW / fs) cos(2 pi F / fs), A = 1 - B - C
effective F_i = F_i * 17.5 cm / L
Peaks of the spectral envelope (formants) distinguish vowels; F0 only sets the harmonic spacing. Reference formants: a 730/1090/2440, e 530/1840/2480, i 270/2290/3010, o 570/840/2410, u 300/870/2240 Hz.