A Note on Autocorrelation Analysis of Speech Sounds

By separating a speech wave into its temporal and structural factors, it is possible to reveal concisely and explicitly how the vocal cavity characteristics and the periodicity of the laryngeal excitation enter into the autocorrelation function of the speech wave. This representation provides a simple explanation for the differences between the conventional spectrogram and the correlatogram, such as the marked sensitivity of the latter to variations in pitch and other nonphonemic qualities of the voice, which have been described in a recent paper by Biddulph. A scheme for using cross correlation to avoid the quadratic suppression of weaker components in the autocorrelation function is suggested.