Live F0, F1, F2 and F3, with medians per take. Hold a vowel for 2–3 seconds, phone about 20–30 cm away, same pitch each time.
Mic processing (echo cancel, noise suppression, auto gain) is switched off, so the full voice comes through.
Recordings from your phone's recorder work too (m4a, mp3, wav).
| Take | F0 | F1 | F2 | F3 | ΔF2 |
|---|
ΔF2 compares each take with the first take of the same vowel. Reference rows are classic adult averages (Peterson & Barney, 1952); people vary a lot around them.
Audio is downsampled to about 11–12 kHz. Each 50 ms frame gets 50 Hz pre-emphasis and a Gaussian window, then Burg LPC (one pole pair per kHz). Formants are the polynomial roots below 5.5 kHz with bandwidth under 1 kHz. Pitch comes from normalized autocorrelation; only clearly voiced frames within 25 dB of the loudest count toward a take.
At higher pitch, F1 on “ee” tends to read 40–60 Hz high, because LPC is pulled toward the nearest harmonic. F2 and F3 stay within about 2%. Compare takes at the same pitch.