Richard J. Baker
Stuart Rosen
Dept. of Phonet. & Linguistics, Univ. College London, 4 Stephenson Way, London NW1 2HE, England
Speech waveforms containing purely temporal information (i.e., flat spectra and instantaneous amplitudes identical to natural speech, obtained by randomly changing the polarity of consecutive samples) contain quasiperiodic fluctuations in amplitude coincidental with voice periods. The temporal complexity of such waveforms can be reduced by half-wave rectifying the speech prior to removal of the spectral information. This study addresses the question of whether such simplification of the waveform improves the transfer of voicing information in a VCV context in normal hearing adults. The results of consonant identification experiments will be presented in which (a) the two polarities of half-wave rectification are compared to the unrectified condition (equivalent to full-wave rectification) and (b) the waveforms are low-pass filtered at the upper extreme of the fundamental frequency distribution of the speaker, prior to full-wave or half-wave rectification and removal of spectral information. Such filtering removes much of the quasi-periodicity arising from voicing in the full-wave rectified condition. [Work supported by the MRC UK.]