These results show that an estimator can be trained on a non-exhaustive data set and can extract physically meaningful sounds from trained data as well as novel sound data. We now go on to generalize this technique to estimating for multiple models simultaneously.