Pedestrians can only safely cross the road before an approaching vehicle if the time remaining until the vehicle arrives at their position (time-to-collision, TTC) is longer than the time needed for crossing. Using a virtual-reality (VR) system that combines physically plausible acoustic simulations of approaching vehicles with visual simulations, we investigated how the vehicle sound affects the perception and behavior of pedestrians in a road-crossing situation. Our results show that 1) for vehicles approaching at constant speed, participants estimate longer TTCs for softer compared to louder vehicles with the same actual TTC, both for auditory-only and audiovisual presentations, indicating potential risks associated with quieter vehicles. 2) When the sound of an accelerating conventional vehicle (ICEV) is presented, this largely removes the inadequate consideration of acceleration observed in visual-only TTC estimation. 3) For electric vehicles (EVs) with and without AVAS, this benefit provided by the car sound is significantly reduced compared to ICEVs. 4) In line with this, the probability of unsafe road-crossing decisions increases significantly with the acceleration level for EVs with and without AVAS, but remains low for ICEVs. Taken together, auditory information is important for pedestrians, particularly so when the approaching vehicle accelerates. Potential risks associated with EVs should be considered.