Representations of Speech Sounds and Some of Their Statistical Properties

Abstract

Spectrographic analysis, autocorrelation, and infinite clipping are considered as methods of transforming and analyzing speech sounds with the object of obtaining simple representations without excessive loss of intelligibility. Although not mathematically equivalent in the exact sense, these methods when properly approximated are shown to provide parameters of statistical nature that are simply related. It is conjectured that the parameters describe some essential elements of speech sounds and are statistically invariant. Some experimental results are included.