Short‐Time “Cepstrum” Pitch Detection

A short‐time cepstrum analyzer for vocal‐pitch detection has been simulated on an IBM‐7094 digital computer. (The cepstrum of a signal is defined as the square of the Fourier transform of the logarithm of the amplitude spectrum of the signal.) Since temporal periodicities in a speech signal cause periodic ripples in the amplitude spectrum, Fourier transformation of the spectrum gives the “frequency” of the ripple, which is inversely proportional to the fundamental frequency of the speech. Taking the logarithm of the amplitude spectrum makes the effects of the vocal tract (spectrum envelope) and the vocal source (spectrum fine structure) additive, thereby separating the low “frequencies” of the spectrum envelope from the usually higher frequencies of the spectrum fine structure. Cepstrum pitch detection is insensitive to phase distortion, amplitude distortion, additive noise, and the absence of the fundamental speech frequency. A general description of the technique and some recent results and examples are...