켑스트럼 파라미터와 다중대역 여기신호를 사용한 음성합성 시스팀

다국어 초록 (Multilingual Abstract)

This thesis deals with a new speech analysis/synthesis method in the viewpoint that improves excitation signals based on a filter of speech synthesis using Multiband Excitation Signals.
The traditional speech synthesis system using analysis/synthesis method have generally used a simple voiced/unvoiced decision, and excitation signals that are made up of white noise signals in case of unvoiced frame, or periodic impluse signals in case of voiced frame. But these excitation signals become the factor which restricts the quality and intelligibility of synthesized speech because it is difficult to generate signals with the spectrum similar to an original speech composed of harmonics and noise signals only by using impluse signals.
This thesis proposes a scheme that improves the intelligibility by restoring noise information mixed in voiced frame with respect to speech quality. First, the voiced/unvoiced decision method for each frequency band in one frame is suggested. It allows the excitation signal for a particular speech segment to be a mixture of periodic (voiced) energy and noise-like (unvoiced) energy. Secondly, the analysis/synthesis system using Multiband Excitation Signals, generated by the information of voiced/unvoiced bands on the spectrum of one frame, is designed in order to synthesize a speech signal similar to the original speech signal.
The result of evaluation shows that this proposed approach can achieved more natural-sounding synthesized speech with less "Buzziness" than the traditional method.

목차 (Table of Contents)

• 목차
• 제1장. 서론 = 1
• 제2장. 한국어 음운에 대한 고찰 = 4
• 2.1 한국어 음소의 이음화 현상 = 4
• 2.2 무성자음의 유성음화 현상에 대한 고찰 = 11
• 2.3 음절수에 따른 유성음화 현상에 대한 고찰 = 18
• 제3장. 분석 시스팀 = 23
• 3.1 기존의 분석시스팀 고찰 = 24
• 3.2 제안하는 분석시스팀 개요 = 25
• 3.3 성도 켑스트럼 파라미터 추정부 = 28
• 3.4 유성/무성 판별부 및 핏치 추정부 = 30
• 3.5 스펙트럼 상의 유성/무성구간 자동 판별부 = 31
• 제4장. 합성 시스팀 = 35
• 4.1 기존의 합성시스팀 고찰 = 36
• 4.1 합성 시스팀 개요 = 37
• 4.2 유성구간 합성부 = 38
• 4.3 무성구간 합성부 = 45
• 제5장. 분석합성 시스팀의 실험 및 결과 = 50
• 5.1 다중대역 여기신호에 의한 음성합성 = 50
• 5.2 핏치주기 변경에 대한 합성음 실험 = 53
• 5.3 실험결과에 대한 합성음 평가 = 55
• 제6장. 결론 = 61
• 참고문헌 = 63