Full sound source localization of binaural signals

A full localisation of binaural signals that comprises of estimating source azimuth and distance under different reverberation time is presented in this paper. The system comprises of two main functional modules. At first, a binaural front-end is constructed for extracting binaural cues, such as interaural time and phase differences, interaural level difference and interaural coherence. A supervised learning of binaural cues such as interaural level difference and interaural phase difference are carried out for localisation in azimuth. The distance estimation is processed by involving all binaural cues from binaural front end for statistical analysis. The distance perception analysis is further improved by integrating the envelope statistical properties of extracted binaural cues with Gaussian Mixture Model-Expectation Maximization (GMM-EM). The developed auditory attention model works effectively without requiring prior knowledge of azimuth position and reverberation time of an enclosed space. The system aims at selection of better ear based on full localisation module to improve the Signal to Noise Ratio (SNR) of the target speaker. The results based on different number of statistical features are tested on different rooms. The equal error rate are computed for full localisation and it is compared with localisation based on only azimuth position.