Extract speech feature vectors for HMM-based Vietnamese speech synthesis system

Phan Thanh Sơn, Dương Tử Cường


Recently, the statistical framework based on Hidden Markov Models (HMMs) plays an important role in the speech synthesis method. The system can be built without requiring a very large speech corpus for training the system. In this method, statistical modeling is applied to learn distributions of context-dependent acoustic vectors extracted from speech signals, each vector contains a suitable parametric representation of one speech frame and Vietnamese phonetic rules to synthesize the speech. The overall performance of the systems is often limited by the accuracy of the underlying speech parameterization and reconstruction method. The method proposed in this paper allows accurate MFCC, F0 and tone extraction and high-quality reconstruction of speech signals assuming Mel Log Spectral Approximation filter. Its suitability for high-quality HMM-based speech synthesis is shown through evaluations subjectively.


Vietnamese speech synthesis, context-dependent, speech parameterization, statistical parametric speech synthesis.

DOI: https://doi.org/10.15625/1813-9663/29/1/2303

Journal of Computer Science and Cybernetics ISSN: 1813-9663

Published by Vietnam Academy of Science and Technology