Enhancing sound sources by use of binaural spatial cues

A highly desirable possible application of simulation of auditory scene analysis is the development of hearing aids capable of suppressing disturbing sound sources in complex noise environments. It is well-known that in such situations spatial cues contribute a small, but often decisive hint for auditory scene analysis. Concerning noise suppression, they have the interesting advantage that they do not require prior knowledge of the sound source spectrum, and that the directions of the sound sources change relatively slowly. Part of these spatial cues are interaural phase and level differences. Using a statistical approach, it is possible to use these parameters from a mixture of sound sources to estimate sound source direction. It has been shown that this works well even in moderate to low signal to noise ratios with an accuracy comparable to what human subjects achieve[10, 8]. Using known head-related transfer functions, estimates of the sound source direction can be directly used to enhance one signal from a two-microphone recording of two signals. However, if more than two sound sources are involved, additional information is needed to estimate the contributions of the different sources. This information can be gained by ”binding” or integrating temporal information from the signal envelope with information on sound source direction. Therefore correlations of direction information and spectro-temporal features are examined. Results show clear correlations between the envelope of the sources and the directional information.