Signal Processing and Speech Communication Laboratory
hometheses & projects › Best Channel Selection for Distant Speech Recognition

Best Channel Selection for Distant Speech Recognition

Master Project
Announcement date
03 Oct 2016
Jonas Helm
Research Areas

Short Description

Conventionally multi-channel microphones acquisition is used for beamforming to reduce interfering sound and reverberation. An alternative is choosing the best channel. This work well if the microphones are not organised as a single array, but as a network of microphones. In recent years several algorithms have been proposed to determine the best microphone channel, that is then used for automatic speech recognition. One approach is e.g. based on the envelope of the speech signal, which gets smeared by reverberation and is there a cue to the quality of the detected signal. A more sophisticated approach uses a frequency dependent analysis, the modulation frequency, just to mention a few.

At the SPSC Lab, we have built a live voice controlled home automation system for Austrian German in our meeting room and kitchen. Microphones are distributed over several walls and the ceiling. We want to have channel selection as an option for our system as a pre-processing step for automatic speech recognition.

Your Tasks

  • Review of recent literature on channel selection for ASR
  • Choose at least one promising approach
  • Real-time implementation of this approach preferably in Python
  • Documentation

Your Profile

  • Motivation and interest in the topic
  • Strong background in (Speech) Signal Processing
  • Experience in Python or Matlab
  • Speech Communication 1 completed


Martin Hagmüller ( or 0316/873 4377)


Martin Wolf, Climent Nadeu, Channel selection measures for multi-microphone speech recognition, Speech Communication, Volume 57,pp 170-180