We train speech models to help AI understand what people say, how they say it, and what they mean—directly from the audio.