HornBase - A Car Horns Dataset

To provide data for machine learning models capable of classifying car horns in a traffic environment, we created HornBase, a dataset comprising 1080 audio files of exactly one-second duration each, manually labeled and balanced into two classes: horn and not horn. The audio excerpts were collected in ten specific situational scenarios in which the audio signals can be received in the car where the horn will be analyzed, as well as three different horn styles: a short honk, a long one, and an intermittent sequence of three consecutive short honks. For each possible audio segment, three temporal windows are cut, with the first containing the initial half of a horn, the second containing the entire horn, and the third containing the final half of a horn.