The present pitch modification methods derive from efforts in our laboratory to develop general prosodic modification capabilities for use in diphone speech synthesis and for enhancing dysarthric speech. The processing method, like PSOLA, must therefore provide independent control of F0 and segment duration. While F0 modification applies only to voiced segments, more general duration control applies to all segments, whether voiced or not. The present method is a hybrid in the sense that it processes voiceless segments (for duration control) using simple time--domain processing similar to the PSOLA method, but uses residual-excited LPC (RELP) processing to reconstruct pitch periods for voiced segment processing. The present report is concerned primarily with the performance of the RELP component of the hybrid LPC method (hereafter HYBRID) in controlling pitch.
Tim Bunnell