適應 ARMA 모형이 적용에 관한 고찰

Abstract

ARMA 모형에 의한 음성 분석에 있어서 가장 곤란점의 하나인 입력추정에 대해서 검토하고 보조수단으로서 임펄스 추정 알고리즘을 제안하였다. 제한된 조건하에서 제 안된 알고리즘은 극히 우수한 특성을 보였으나 조건의 변화에 따라 그 특성이 급격히 열화되는 것이 관측되었으며 일반적 적용은 타당하지 않음이 판명되었다. 아울러 파라메터 추적과정에서 수렴에 적절한 신호의 진폭과 수렴판정의 척도에 대해서 평가와 제안을 실었다.

The major difficulty in analyzing the voice signal by ARMA model lies on the fact that the excitation signal can not be separated from the estimation error. An algorithm appropriate to estimate isolated impulse type excitation signal is presented along with the brief discussion of recent work related to the topic.

The proposed algorithm exhibits highly reliable performance under the favorable conditions. However it has been found the algorithm is so sensitive to the change of condition that it's applicability is limited to the special purposes. Also, the effect of the signal amplitude on the convergence, and the measure of convergence are discussed, and suggestions are given for better use of ARMA model for voice signal analysis.

The major difficulty in analyzing the voice signal by ARMA model lies on the fact that the excitation signal can not be separated from the estimation error. An algorithm appropriate to estimate isolated impulse type excitation signal is presented along with the brief discussion of recent work related to the topic.

The proposed algorithm exhibits highly reliable performance under the favorable conditions. However it has been found the algorithm is so sensitive to the change of condition that it's applicability is limited to the special purposes. Also, the effect of the signal amplitude on the convergence, and the measure of convergence are discussed, and suggestions are given for better use of ARMA model for voice signal analysis.