A widely used technique for the representation of sensor data is based on diagonalizing the correlation tensor of the data-set, keeping a small number of coherent structures (eigenvectors) based on principal components analysis (PCA). This approach tends to be global in character. It is possible to combine wavelet analysis and PCA to obtain proper accounting of global contributions to signal energy without loss of information on key local features. We plan to exploit such a combined wavelet-PCA technique in auditory processing, adapting it to different types of data domains and processing constraints by effective choices of wavelet bases. Our approach starts from the wavelet coefficient image in the phase plane. The very choice of analyzing wavelet (and hence the basis) leads to highlighting of certain features through the strengthening of a small set of coefficients, leaving the remainder at low amplitudes. Given an ensemble of such images, one can perform pattern extraction from the ensemble using a Karhunen-Loeve decomposition also known as PCA. The components are then used as a compact code. In recent work of Krishnaprasad and collaborators [.Kugarajah 1996.], this idea is used to extract patterns from reflectance spectroscopy. We propose to investigate adaptive wavelet-PCA approaches to pattern extraction from auditory data, signal inversion and feature enhancement in the presence of noise. We also plan to investigate fast parallel implementations of this approach.