Real‐time spectral analysis

AbstractFor signal analysis, the Fourier transform and the linear prediction methods are the main approaches. In recent years, the wavelet transform, which gives an expression in the time–frequency domain, has been attracting attention. In the wavelet transform, scale and shift operations are used to analyze regional waves in the signal. In this paper, we propose a new method for signal analysis. Instead of the wavelet, short sine waves and cosine waves which have slightly different frequencies and phases established beforehand are used. These frequencies and phases play the same role as wavelets in analyzing regional waves in the signal. However, our proposed method is different from the wavelet transform because it does not use multiresolution analysis. In conventional spectral analysis methods, all data are prepared beforehand and calculations are performed by using these data. In the proposed method, the spectrum is updated whenever new signal data come in. We show that the amount of calculation is markedly reduced and that real‐time spectral analysis can be achieved. Finally, we present an analysis of Japanese vocal sounds using the proposed method. © 2007 Wiley Periodicals, I