HMM 합성 기술과 마이크로폰 어레이 기반 잡음 제거 기술에 관한 연구

Abstract

HMM-based speech synthesis system models the speech parameters as the context-dependent HMMs. It is important to estimate correctly the MVF in the two-band excitation model because it is one of factors deciding speech quality. In this paper, we describe the HMM-based Korean speech synthesis system and propose the re-estimation method of MVF to estimate more correct MVF. The proposed method produces better quality-speech. And, we describe and analyze the performance of adaptive beamforming algorithms which are GSC -based CCAF, Frost algorithms, the filter bank- based CCAF and filter bank-based Frost algorithm. For the performance evaluation, the SNR and BSD are measured. Filter bank-based frost algorithm shows the best performance.