Turkish Text-To-Speech Synthesis For A Set Of Sentences Using Prototype Waveform Interpolation Method

摘要

In this study, the application of concatenative Text-to-Speech Synthesis (TTS) using Prototype Waveform Interpolation (PWI) technique was investigated for Turkish. For the experimental research, diphones have been used as the database units and Turkish diphone database which has been prepared for MBROLA (MultiBand Resynthesis Overlap Add) diphone synthesiser has been used. The feasibility of PWI technique to text-to-speech synthesis was investigated, and during diphone transition areas, pitch period and waveform interpolation has been done in order to minimize the mismatches at the concatenation points. The implementation of a Natural Language Processing module and prosodic modifications are not within the scope of this study and therefore the proposed system is not a high-quality TTS. However the performance evaluation that has been done at the end of the study revealed that PWI technique can be used successfully to smooth transitions at the speech segments' in order to improve the synthesised speech quality.