UTDrive Microphone Array

1
UTDrive Microphone Array
Kevin Derischebourg Chance Kelly Taehwan Kim
Jeffrey Leinbaugh Sean Stixrud
kcd031000_at_utdallas.edu
ckellyusa_at_gmail.com
thk061000_at_utdallas.edu
jeff.leinbaugh_at_gmail.com
sfs041000_at_utdallas.edu
Department of
Electrical Engineering University of Texas at
Dallas Erik Jonsson School of Engineering
Computer Science Richardson, Texas
75083-0688, U.S.A.

New
Old
Microphone Arrays mounted in the car
Before POWER CIRCUITRY After
Organization
Speaker Recognition

• Preprocessing steps
• Original audio.
• Filtered audio.
• Voice Activity Detection (VAD), speech starts and
  stops.
• Fully processed audio with speech only.

Results
Conclusion

• Speaker recognition was performed with
  unprocessed audio and beamformed audio.
  Beamformed audio outperformed the unprocessed
  audio, resulting in twice as many correct matches
  using significantly shorter test samples.
• The array configurations performed similarly in
  this experiment, while the CSA-BF method
  outperformed DASB by a 3.2dB margin.
• The best performing setup in this experiment was
  CSA-BF paired with either a linear or logarithmic
  array configuration, which provided a 4.4dB
  segSNR improvement over a single channel.