Description:
Capturing the sound from the audience during broadcasting and/or recording lectures presents challenges due to increased distance between the microphone(s) and the speaker which worsens the signal-to-noise-ratio and increases the reverberation in the captured signal. Using microphone array and beamforming algorithms substantially improves the quality of the captured sound. Reliable and precise sound source localization (SSL) is critical for pointing the beam to the right direction to capture speech. We investigate potential techniques for SSL in this scenario. A scalable two dimensional SSL algorithm is proposed based on the space of instantaneous direction of arrival (IDOA). Due to reverberation and noise effects, the estimates of IDOA per frame per frequency bin are not robust enough. To counter the adverse effects, several weighting and filtering techniques are applied to process and combine IDOA to obtain more reliable estimates of speaker locations suitable for practical applications.
Speaker(s):
Rong Hu, fourth year PhD student, Department of Computer Science, University of Missouri-Columbia
|
