Audio Signal Analysis

Audio Signal Analysis is the basis of most audio related research - a close collaboration with other AudioLabs research areas is therefore natural. One of the fields addressed in the group headed by Prof. Dr. Bernd Edler is the development of improved models of auditory perception suitable for audio codecs as well as for speech codecs. Besides new spectral analysis systems and extended tonality estimation, different control methods for speech and audio codecs are studied. These developments are accompanied by psychoacoustic experiments for the systematic evaluation of specific auditory phenomena.

Time/Frequency analysis is a ubiquitous element of contemporary digital signal processing algorithms. The main research topics in this field encompass real- and complex-valued transforms and time-warped filter banks as well as their temporal and spectral adaptation to the characteristics of the input signal. Applications already investigated include perceptual audio coding, parametric signal representations as well as analysis and enhancement of both speech and music.

Most currently available audio signals are professional mixes of multiple sound sources. The capability to modify (e.g. attenuate, amplify) sound sources isolated from the mixture enables new applications with respect to efficient source coding and interactive services. If the mixture is the only available representation (blind source separation) appropriate signal representations are important. These representations must reveal the different characteristics of the individual sources and enable access to their unique features, e.g. timbre and modulations.