Summary and Future Directions

We presented a framework in which we can estimate which two of $N$ sources are active on a stereo recording by using DASSS data at a given point in time-frequency space. We outlined a Bayesian structure to make our hypotheses. In it, the data $D$ is either the single-active-source parameter estimates of the DUET system or data obtained using delay and scale subtraction scoring (DASSS). In the former case, computational expense is a currently prohibitive factor. In the latter, we demonstrated an application of the Bayesian approach. In the future, we will turn to the rest of the Bayesian expression, developing rubrics for choosing $P(u,v,r)$ to capture source knowledge, and to exploiting and representing DASSS data more efficiently.