An experimental study of energy dips for speech and music

摘要

Speech signals generally have many intervals with low energy “energy dips”. For music signals, energy dips are not always remarkable. We studied stochastic features of energy dips for speech and music signals. A certain difference was found between the signals in the length of energy dips. Also, the number of energy dips in a time window and their distribution were investigated. From this distribution, a threshold number of energy dips was estimated which provided a scheme for the discrimination of speech from music.