ABSTRACT
There may be instances where some sources may be specified to be observable while others are not. An example of such a situation is when we recognize that a stereo at a cocktail party is tuned to a particular radio station. We record the radio station in isolation; thus the source is said to be observable, but the associated mixing coefficients for this source are still unknown. The following model allows for such situations. The following model is a combination of the Bayesian Regression model for observable sources introduced in Chapter 8 and the Bayesian Source Separation model for unobservable sources described in Chapter 10. Either model may be obtained as a special case by setting either the number unobservable or observable sources to be zero.
