Binaural hearing computer models in multisource environments

Computer models of the binaural hearing mechanism are constantly developed and improved in order to simulate the hearing process as good as possible. This is necessary because of increasing need to localize and hear a certain speaker among others, simultaneously speaking sound sources, especially in teleconferences or for hearing-impaired people. The basics of almost all computer models is the cross-correlation function of the binaural signal, adequate filtered and processed. The final goal is to filter out from the binaural signal a certain speaker without any noise or interference. This paper uses a complex short-time cross-correlation model to find its limitation in identifying the right number of sound sources in the frontal horizontal plane, as well as their position (azimuth). The tests were done using HRTF functions to simulate a virtual acoustic environment without reflections. The results were compared with tests made in real rooms using dummy head for binaural signal recording.