A pitch estimation algorithm for speech and music

We present an efficient pitch estimation algorithm which operates in the time domain by comparing features extracted from the signal being analyzed. The algorithm uses both primary features (extracted directly from the signal) and secondary features (derived by combining several primary features to reduce the effect of variations in the signal envelope). We relate the algorithm both analytically and quantitatively to autocorrelation analysis. The algorithm is shown to function successfully for both speech and music over a six octave pitch range, and to be appropriate for operation in real time.