For best results, select a recording with overlapping or background speech to observe how the model performs.