Synthesis by Concatenation of Waveform Segment

• Main Authors: Shiou-Ming Chu, 朱修明
• Other Authors: Ching-Tang Hsieh
• Format: Others
• Language: zh-TW
• Published: 1994
• Online Access: http://ndltd.ncl.edu.tw/handle/61073183594518265938

碩士 === 淡江大學 === 資訊工程研究所 === 82 === In this paper, we devote to synthesize Chinese word byans of waveform splicing base on phoneme and pitch period. To synthesiae speech by concatenation of phoneme or pitchriod naturally and fluently, there are two important thingsconcemed : 1. How to capture the fundamental waveform. 2.handle the situations when splicing. About these problems, we provide some processes. Theirprovement in quilities of synthesized speech was confirmed byriments: (1) How to capture the fundamental waveform: The Quility of synthesized speech is highly dependent on the selection method of speech waveform while building speech database. There are two types of speech signal we want toapture: consonants and vowels. 1. When collecting consonant: we had retain pure consonant part and some periods of vowel which adjoin to. In this way, we can make the CV type concatenation easily and fluently. 2. When collecting vowel: using FFT Ceptrum to estimate pitch of speech segament. We can find a maximum amplitude (MA) in the pitch period. Finding the nearest zero-crossing point before MA as a starting point of pitch period and regarding the previous point of next period' srting point as ending point. Repeat the capturess, we retain nine pitch periods in the range of 3. The high sampling rate is applied to build speech database. The higher resolution of speech waveform can reduce the distortion of interpolation process. Thus the synthesized speech with higher sample-rate database islosre to original speech than the lower one. (2) How to handle the situations when splicing: The concatenation problem take place in these situations: 1. Concatenation of consonant and vowel (CV type):Concatenation of pitch periods in a vowel (Vtype):