Vehicle speed is an important factor influencing highway traffic safety. Radars are applied to control the speed of vehicles, but the drivers often decelerate when approaching radar, and then accelerate after passing by. We address automatic recognition of speed change from audio data, based on recordings taken in controlled conditions. Data description and classification experiments illustrate both changing speed and maintaining constant speed. This is a starting point to investigate what percentage of drivers actually maintain constant speed, or slow down only to speed up immediately afterwards. Automatic classification and building an appropriate database can help improving traffic safety.