Signal Processing and Speech Communication Laboratory
hometheses & projects › Multi-Channel Speech Enhancement

Multi-Channel Speech Enhancement

Master Thesis
Announcement date
17 Feb 2012
Lukas Pfeifenberger
Research Areas

Today, speech transmission systems like telecommunication devices or human-machine interfaces can be found almost everywhere. As a consequence, these systems are exposed to all kinds of environmental noise. Especially for low SNRs, speech quality and intelligibility are severely degraded. The purpose of speech enhancement systems is to lessen these influences by the means of digital signal processing.

For applications where only a single microphone is available, myriads of speech enhancement algorithms have been conceived. Most of them are based on the temporal di erences in the statistics of human speech and the interfering noise. However, all of these methods require a trade-o ff between noise cancellation and speech distortion. In the last decade embedded systems have become more powerful and aff ordable. Therefore, multi-microphone approaches for speech enhancement gained an increasing interest in industrial and commercial applications. When more than one microphone is present, additional information becomes available. Next to the temporal information also the spatial features of the sound field can be exploited. A beamforming array can be thought of a spatial lter, which achieves speech enhancement by attenuating signals which do not originate from the same direction as the desired speech signal. However, spatial fi ltering alone cannot deliver sufficient
noise suppression especially in diff use noise fi elds. Therefore, a post-processing stage is used to further enhance the beamformer output. This post filter is often a common single-channel noise canceling algorithm, which solely relies on the signal statistics. Recently, these algorithms have been adapted to also bene fit from the spatial information obtained by the array.

Profile of prospective student

The candidate should be interested in machine learning, applied mathematics/statistics, Matlab programming, and algorithms. Interested candidates are encouraged to ask for further information.