Estimation of the Probability Density Function of the Interaural Level Diferences for Binaural Speech Separation

Source separation techniques are applied to audio signals to separate several sources from one mixture. One important challenge of speech processing is noise suppression and several methods have been proposed. However, in some applications like hearing aids, we are not interested just in removing noise from speech but amplifying speech and attenuating noise. A novel method based on the estimation of the Power Density Function of the Interaural Level Differences in conjunction with time-frequency decomposition and binary masking is applied to speech-noise mixtures in order to obtain both signals separately. Results show how both signal are clearly separated and the method entails low computational cost, so it could be implemented in a real-time environment, such as a hearing aid device.