Mistuned voices destroy phase predictability and broaden perceived pitch

Evidence

If a soloist holds a 220 Hz note with no vibrato, the sound and the autocorrelation is perfectly periodic. Not so for a chorus: a somewhat random addition of sinusoids takes place, one for each partial of each singer’s voice as crests and troughs from different singers drift in and out of phase since they are not perfectly in tune. A given peak in the signal, the sum of all singers, may not be at all predictive of where the peaks in the signal lie a quarter-second later, implying that the autocorrelation decays within that quarter-second (or perhaps much faster). The autocorrelation decay on a timescale Δt broadens the power spectrum by Δf, such that Δf Δt ~ 1, in accord with the uncertainty principle. — Eric J. Heller, Why You Hear What You Hear- An…, 420

Categorization

• XREF: Directly explains why live choirs sound different from solo recordings; connects to the Fourier/certainty principle already invoked earlier in the book.
• DEFINE: Explains the mechanism linking autocorrelation decay to power spectrum broadening via Δf Δt ~ 1.