바이스펙트럼 및 신경회로망을 이용한 음향인식

Abstract

어떤 신호를 주파수영역에서 해석하고자 할 때 바이스펙트럼 기법은 파워스펙트럼 기법에 비하여 상대적으로 계산 시간이 많이 걸린다는 단점이 있으나 신호의 크기에 관한 정보 뿐 만 아니라 위상에 관한 정보도 찾아 낼 수 있다는 장점이 있다. 한편 바이스펙트럼은 1차원 함수인 파워스펙트럼과 달리 2차원 함수이므로 등고선 그림으로 표시된 바이스펙트럼은 하나의 무늬로 볼 수 있다. 측정된 음향신호의 바이스펙트럼을 무늬로 표시하면 각 음향신호에 대한 바이스펙트럼 무늬는 서로 다르다. 따라서 음향 신호원을 식별하기 위한 한가지 방법으로 각 음향신호의 바이스펙트럼 무늬를 식별하여 궁극적으로는 음향신호원을 식별하는 방법을 제안하였다. 본 논문에서는 여러 종류의 무한궤도차량(無限軌道車輛)의 음향신호를 측정하여 음향신호의 바이스펙트럼을 각각 구한 후, 신경회로망을 이용하여 각 바이스펙트럼을 식별하여 결과적으로 음향신호원을 식별하는 방법을 보였다.

Bispectrum has the magnitude information of a signal in the frequency domain as well as the phase information of the signal. However, it takes longer time to compute the bispectrum of the signal than to compute the power spectrum which has only the magnitude information of the signal. Although the power spectrum is a 1-dimensional function, the bispectrum is a 2-dimensional function so that the contour plot of the bispectrum can be considered as a pattern. Since the patterns of the bispectra of the measured acoustic signals are different each other, a pattern recognition technique is proposed to recognizing the bispectra of the acoustic signals as a method to identify the sources of the acoustic signals. In this paper, the bispectra of the acoustic signals which are measured from several kinds of caterpillar vehicles are computed, then a neural network is used as the identifier of the bispectrum patterns of the caterpillar vehicles, as a result, the bispectrum pattrns are used to identify the sources of the acoustic signals.

Bispectrum has the magnitude information of a signal in the frequency domain as well as the phase information of the signal. However, it takes longer time to compute the bispectrum of the signal than to compute the power spectrum which has only the magnitude information of the signal. Although the power spectrum is a 1-dimensional function, the bispectrum is a 2-dimensional function so that the contour plot of the bispectrum can be considered as a pattern. Since the patterns of the bispectra of the measured acoustic signals are different each other, a pattern recognition technique is proposed to recognizing the bispectra of the acoustic signals as a method to identify the sources of the acoustic signals. In this paper, the bispectra of the acoustic signals which are measured from several kinds of caterpillar vehicles are computed, then a neural network is used as the identifier of the bispectrum patterns of the caterpillar vehicles, as a result, the bispectrum pattrns are used to identify the sources of the acoustic signals.