Ideally without changing interface too much. Want to be able to specify at train time which datasets to use, and what proportions of each to use. For example, if we wanted to train on 80% of the earbud dataset, and 50% of the adolescent dataset, to understand how much synthetic data versus real vocal call data we'll need.
Ideally without changing interface too much. Want to be able to specify at train time which datasets to use, and what proportions of each to use. For example, if we wanted to train on 80% of the earbud dataset, and 50% of the adolescent dataset, to understand how much synthetic data versus real vocal call data we'll need.