This page lists some of the projects I’ve been working on.
To allow for an unsupervised restoration of impulsive disturbances in archive audio applications, we have developed an algorithm that classifies an audio recording as either disturbed or undisturbed. The corresponding website contains more information and makes a number of demonstration signals available for listening.
Most of the existing algorithms for noise power spectral density (PSD) estimation in audio signals were developed for the application with speech signals. As a consequence, large estimation errors may occur when these algorithms are applied to signals different from speech, e.g., music signals. To make unsupervised restoration of large amounts of diverse audio material possible, we have developed a noise PSD estimation algorithm that yields accurate estimation results for a large diversity of input signals. The corresponding website contains more information and makes a number of demonstration signals available for listening.