Loading...

Speaker Localization in Noisy Environments: Design and Implementation of a Robot Auditory System

Saffari, Ehsan | 2016

651 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 48209 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Meghdari, Ali; VazirNezhad, Bahram; Alemi, Minoo
  7. Abstract:
  8. Natural intuitive human-robot interaction depends on how responsive a social robot can be to environmental information. Speaker localization is a prominent ability for communication and socialization that allows the robot to react to a speaker by turning its head towards it. In this paper, we have designed and implemented a robotic head, “Ava”, which turns toward the speaker in noisy environments. Ava is able to detect and enhance a speech activity and estimate the azimuth direction and vertical level of a speaker. The speech segments which are differentiated by a Speech Activity Detection system are processed to reduce noise levels. Then, by extracting the binaural features of two signals recorded by Ava’s microphones, the speaker is localized. Finally, Ava turns toward the speaker in a smooth path and is able to update the estimation even while the head is turning. So all of the effects of robot rotation on this estimation are studied. As Ava is an unknown audition system, we used a two-layer perceptron neural network for sound source localization. We proposed that Ava can turn toward any speaker in noisy environments with SNR of 20 dB with the precision of +/-10 degrees in horizontal plane. In addition, in the same condition, vertical directions of speaker are classified to three levels with the accuracy of 90%
  9. Keywords:
  10. Social Robotics ; Speech Processing ; Human Robot Interaction (HRI) ; Speaker Localization ; Ava Robot

 Digital Object List

 Bookmark

No TOC