Autocorrelation detects pitch by finding the lag at which a signal most resembles itself

Autocorrelation computes the similarity of a signal with a delayed copy of itself across a range of lags. The lag at which similarity peaks corresponds to the period of the fundamental frequency: period T → pitch = sampleRate / T. Unlike FFT, autocorrelation works in the time domain and is robust to harmonic-rich signals where the fundamental may not be the loudest partial. The resulting correlogram — a plot of correlation vs. lag — shows a peak at the fundamental period and smaller peaks at harmonics. Center clipping (zeroing samples below a threshold) before autocorrelation reduces noise and suppresses false peaks from aperiodic components.

Examples

Autocorrelate Time Domain and Autocorrelation Circle demos show the correlogram visualization; Pitch Track demo maps the peak lag to a displayed frequency value.

Assessment

Explain why autocorrelation finds pitch even when the fundamental frequency is weak in the spectrum; describe how center clipping improves detection in noisy signals.