Methods for Co-channel Speaker Separation

Co-channel speaker interference occurs when the voices of two or more speakers are superimposed on a single communications channel. This type of interference may occur due to noisy background environments at the transmitter site, or the interference may be introduced in the communications channel itself, such as when multiple radio transmitters are transmitting at the same radio frequency. Regardless of the cause, the resulting co-channel signal received at the other end of the communications channel is often highly unintelligible for human listeners. Furthermore, such co-channel interference usually results in drastic degradations in performance of automatic speech processing systems, such as speech recognition and speaker identification systems. It is, therefore, desirable to separate, or otherwise recover from the co-channel signal, the speech signals of each of the individual speakers whose voices are present in the co-channel signal. This task is often referred to as co-channel speaker separation. The goal of the proposed Phase I research is to investigate and develop the key processing units of a system capable of performing co-channel speaker separation. This will be accomplished by 1) developing a new method for performing speech separation on individual frames of co-channel speech, and 2) developing a new method for determining the number of voices present in a given frame of co-channel speech.