Objective measures of peak clipping and threshold crossings in continuous speech

This study reports data on the statistics of instantaneous speech levels in continuous speech samples, with special emphasis on threshold crossings and other quantities related to peak clipping. All clipping thresholds for each speech sample are defined with respect to the individual speech level for that sample, specified in equivalent peak level (epl). Speech clipping is also treated as speech-correlated noise by assuming that it is caused not by a voltage limiting process, as actually occurs, but by an additive “phantom signal” that will cause the original signal to appear to be clipped. Empirical measures are obtained for the percent time a clipping level is exceeded, for the relation between phantom signal (i.e., noise) power and clipping level, and for the loss in signal power resulting from clipping. The relation is also established between epl and average (rms) power to allow signal power and signal-to-clipping noise levels to be specified.