Analysis of Vietnamese tones to optimize database in speech synthesis using unit selection method

This paper presents a novel approach to optimize data in Vietnameses speech synthesis using Unit Selection method. First, we conduct analysis of Vietnamese tone using Fujisaki model to find out the parameters of fundamental frequency contours (F0 contours) influencing on Vietnamese vowels while speech is expressed. Next, analysis, testing, and evaluation of the effects on the vowel are performed. After that, the data of unit selection speech synthesis system is optimized by recording the vowel with a level tone. As a result, when a Vietnamese word is synthesized, it will be synthesized with a level vowel first followed by being adjusted the parameter of F0 contour to create a word with appropriate tone. With this approach, recorded data can be reduced up to 64,44% while sound quality is insignificantly affected.