Signal combining for distributed microphones

• Applicant: Professor Dr.-Ing. Jürgen Freudenberger
• Subject Area: Acoustics ; Electronic Semiconductors, Components and Circuits, Integrated Systems, Sensor Technology, Theoretical Electrical Engineering
• Term: from 2011 to 2016
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 193880614

Although microphone beamforming has many advantages over voice recording with one microphone, classic microphone arrangements and methods have their limitations in many applications. Such arrangements with microphones that are mounted in a small distance are problematic, especially in situations where the speaker is relatively far away from the microphone array. In such applications the suppression of noise and reverberation can often be significantly improved by increasing the distance of the microphones. Such distributed arrays typically result in very different signal conditions at the individual microphones. This diversity can be exploited for robust signal enhancement, but leads also to new issues for the signal combining.The aim of this project is the development of signal combining methods for distributed microphone arrays. Such algorithms should extract one or several desired speech signals from the microphone signals, which are disturbed by noise and reverberation. Applications are, for example, handsfree devices in vehicles or telephone conferencing systems. In the first phase of the project, it was shown that diversity techniques can be applied to speech signals profitably. The algorithms developed in preliminary work have been extended to situations with multiple speakers and noise fields with correlated disturbances. Furthermore, other known combining methods such as the Multi-Channel Wiener Filter (MWF) and Relative Transfer Function Beamforming have been investigated and adapted for distributed microphone arrangements. All developed procedures require only an estimate of the correlation properties of the input signals.In this application an extension of the project for another 24 months is requested. The scope of this second project phase is to enhance the signal combination and parameter estimation, in particular in scenarios with multiple speakers. Further open issues regard the choice of a reference channel. Several combining methods, for instance the MWF, choose a reference channel, which significantly impacts the signal quality of the output signal. The diversity combining reduced this dependence in comparison to other methods. Nevertheless, the proposed diversity combining depends on the choice of a reference phase. This dependency should be dissolved or reduced. Other questions regard the parameter estimation and signal combining for multi-speaker scenarios.

• DFG Programme: Research Grants