Generative models are used to create synthetic training data, such as a voice pronouncing a name with simulated background noise like a baby crying