Speech synthesis by dyadic interpolation of spectral parameters

Abstract

Recent work of Olive and Spickenagel on dyadic interpolation of pseudoarea parameters derived from linear prediction analysis has shown that speech synthesized from these parameters suffers very little distortion due to the dyadic interpolation. This raises a question: Can other acoustic parameters, such as the power spectrum of the speech waveform, be similarly interpolated? The spectrum is of special interest because speech can be synthesized in real time from spectral parameters on a readily available programmable digital filter. To study this question a speech analysis-synthesis system using spectral parameters (samples of power spectra at different frequencies) was simulated. These parameters were determined from the speech signal at every dyad boundary, and interpolated for intermediate values. Dyad boundaries (representing the limits of transition regions between phonemes) were determined interactively. Informal listening tests comparing synthetic speech with and without dyadic smoothing showed slight degradation in the smoothed speech. The degradation results from nonlinear variations of spectral parameters within some dyads. Methods for handling these nonlinearities will be discussed. The results of our study suggest that dyadic interpolation works as well for spectral as for area parameters.