Abstract:
Line spectrum pair (LSP) is one representation of linear predictive coding coefficients (LPC coefficients) which is used in formant coding of speech signal. It provides the stability on the interpolated parameters. This thesis proposes a Thai syllabic speech synthesizing method and its phonemes. This method uses the property of the line spectrum pair to encode phonemes and to generate formant transition between phonemes using linear interpolation. These units contain formant locus of Thai phonemes. In the synthesis method, the data are analyzed from speech signal and are used as a synthesizing database. These data consist of fundamental frequency patterns which are used in regenerating tone; amplitude envelopes which are used in controlling the amplitude envelope of synthesized speech; and time duration which is used in controlling the duration of the synthesized phoneme in each synthesized syllable. To synthesize Thai syllables, the thesis proposes a method of synthesizing speech from units of different syllabic structures. The units are classified into different types of segments by their functions and sounds, then the linear interpolation of line spectrum pairs is operated to generate speech from these units. To regenerate tone, the TD-PSOLA method is selected to implement this work. By this synthesis method, all Thai syllables can be synthesized. The speech quality of this synthesis method was assessed by 10 volunteers. The results of speech quality assessments have MRT scores of 78% and MOS of 3.98