Pitch Marking Study towards High Quality in Concatenative Based Speech Synthesis

Abstract

Generally, concatenative speech synthesis systems provide a considerable synthesis quality since the criteria for unit selection methods have been optimized. However, the level of synthesis quality still depends on the adequate concatenation of speech units. An adequate concatenation of speech units has as precondition that concatenation mismatches as phase mismatch, phase mismatch and discontinuity of spectral envelope must not appear in the synthesized speech signal. Therefore, avoiding phase mismatches leads to a high speech synthesis quality and the way to avoid phase mismatches is achieved by an appropriated pitch marking algorithm. Therefore, a pitch marking study was carried out through a evaluat-ing the available pitch marking algorithms. So, a speech database was pitch marked many times using the different pitch mark algorithms. Therewith, several sentences were synthesized applying the different pitch makings of the speech database. A mean Opinion Score (MOS) listening test was carried out for the evaluation of the synthesized speech sentences regarding mismatch human perception. The best pitch mark algorithm was selected according its observed effect in the quality of the speech synthesis.