FarsBayan: A Unit Selection based Farsi Speech Synthesizer

In recent years, the unit selection-based concatenative speech synthesis method using a large corpus has attracted great attention. This method provides more natural quality speech compared to the parameter driven methods. The Formant Synthesis, HNM method and use of MLSA filter are the prevalent methods for synthesizing Farsi speech. In this paper, we present the structure of a proposed unit selection synthesizer for Farsi language. In the proposed system, the linear regression method has been used for determination of weights of discrete sub-costs in the target cost, while the weights of other sub-costs have been considered constant. We have also presented a pre-selection algorithm using adaptive threshold for pruning the units. In addition, the efficiency of TD-PSOLA algorithm in improvement of resulting speech quality has been studied. Informal tests show the degrading effect of this algorithm on the output quality. The output speech was found to be remarkably fluent and natural. The quality of the output speech has been evaluated using MOS subjective test, and we have obtained a MOS test value of 3.8 for overall quality.