Second International Conference on Spoken Language Processing (ICSLP'92)
Banff, Alberta, Canada
This paper describes a series of perceptual tests which demonstrate that listeners are capable of perceiving the systematic cycle-by-cycle changes which occur in the voice source waveform of natural speech as a result of laryngeal coarticulation. Having established that these dynamic changes are perceptible in natural speech we attempt to replicate the effect in synthesis using the KLSYN88 software speech synthesiser. This confirms that our synthesis strategy is appropriate and highlights the most important features of the glottal flow waveform needed to model the changes found in natural excitation. We then turn to detailed analysis of the anticipatory and perseverative coarticulation effects on vowels in the context of British English alveolar obstruents for male and female speakers by means of inverse filtering. The findings demonstrate predictable trends in the voice source parameters associated with different phonetic and allophonic variations and are allowing us to develop rules for the voiced excitation in these phonetic environments for high quality text-to-speech synthesis.
Bibliographic reference. Palmer, Sarah K. / House, Jill (1992): "Dynamic voice source changes in natural and synthetic speech", In ICSLP-1992, 129-132.