Detection of Abrupt Changes in Sound Signals with the help of Wavelet Transforms

Authors: Grossmann A., Holschneider M., Kronland-Martinet R., Morlet J.
Publication Date: January 1987
Journal: Advances in Electronics and Electron Physics (Suppl. 19, Inverse problems, pp 289-306, Academic Press)


So-called “wavelet” or (“ondelette”) methods have been used recently with success in a variety of fields. The present paper, which follows a related one (Kronland-Martinet R. et al. “Analysis of sound patterns through wavelet transforms” Int. Journal of Pattern Recognition and Artificial Intelligence, vol. 1, pp.273-302, 1987) is mainly concerned with sound signals. The “wavelet” techniques are here applied to a discussion of questions like : “When did “something new” appear in the signal ?” and : “How does the new component behave in a neigbourhood of its arrival time ?”. Roughly speaking, a convenient answer to the first question is given by the lines of constant phase of the wavelet transform, which tend to converge towards the points of interest ; the second question is answered by the characteristic behaviour of the modulus of the wavelet transform. Our discussion will start with some basic theory ; we continue with the analysis of computer-generated signals with known threshold behaviour, in the presence of computer-generated noise. We end with some speech and scratchy music. This paper was written with the aim of illustrating some applications of wavelet transforms. We don’t feel competent to discuss in general the topic of abrupt changes in signals.