A new machine-learning model learns to pinpoint exactly where a particular sound occurs in a video clip without the need for human intervention. The model could have applications in areas like journalism and film production or education and training.