Binaural CASA algorithm for speech source localization: Advancements in noisy and reverberant situations

Binaural CASA algorithm for speech source localization: Advancements in noisy and reverberant situations

Opdam, R.C.G.

Schlesinger, A. (mentor)
Boone, M.M. (mentor)

Applied Sciences

Imaging Science & Technology

Laboratory for Acoustical Imaging and Sound Control

2010-11-29

In this thesis a binaural CASA localization algorithm is developed for the implementation in a binaural hearing aid with downstream speech enhancement. Two binaural CASA localization algorithms, based on the Albani model, are proposed to enhance the localization performance in noisy and reverberant acoustic environments. The Albani model is extended with a zero-lag interaural coherence (IC) time window pre-selection, detection of multiple sources per time-window, coincidence detection between interaural level and time differences (ILD and ITD) and a lagged time window comparison, in the proposed extended Albani algorithm. A further addition to the proposed extended Albani algorithm with a binaural cue selector based on an inhibition process, is proposed in the extended Albani algorithm with cue selection by inhibition. Performed simulations show that the extended Albani algorithm performs the best in noisy situations up to a SNR level of -12 dB and the extended Albani algorithm with cue selection by inhibition performs the best in reverberant situations up to a reverberation time of 2.0 s. These proposed localization algorithms show a better performance than the present known CASA methods in both noise and reverberation.

Source localization
CASA
Binaural cue
Hearing aid
Speech intelligibility
Cocktail party problem
Inhibition
ILD
ITD
Noise
Reverberation
Albani

http://resolver.tudelft.nl/uuid:7b397f85-9573-4940-892f-bcf9ae739545

Student theses

master thesis

(c) 2010 Opdam, R.C.G.