Second ESCA/IEEE Workshop on Speech Synthesis September 12-15, 1994 Mohonk Mountain House, New Paltz, NY, USA

Rule-Based Female Speech Synthesis - Segmental Level Improvements

Inger Karlsson, Lennart Neovius

Dept of Speech Communication and Music Acoustics, KTH, Stockholm, Sweden

In this paper we will discuss development of rule-based female speech synthesis, focusing on segmental level improvements. A higher degree of naturalness is achieved by the inclusion of voice source variations and more complex articulation modelling. Source, formant and pole/zero parameter data generated by inverse filtering have been interpreted and incorporated into the female text-to-speech rule system. In particular, the quality of consonants with complex articulations and vowel-consonant transitions is enhanced. A diagnostic intelligibility test was performed on the preliminary rule system. Results are discussed in terms of error sources and possible solutions. On-going work includes fine-tuning of LF-source and pole/zero parameter control, based on these results.

Full Paper

Bibliographic reference.  Karlsson, Inger / Neovius, Lennart (1994): "Rule-based female speech synthesis - segmental level improvements", In SSW2-1994, 123-126.