Predicting EV Warning Sounds

Electric vehicles (EVs) are so quiet at low speeds that they can be a danger to pedestrians, cyclists, and other road users. So the challenge for car manufacturers is to create sounds that are detectable but not annoying.

To avoid them becoming a hazard, the proposed legislation requires EVs to emit warning sounds at low speeds. For consistent evaluation, an objective algorithm is needed to predict how quickly subjects can detect the sounds and their perceived annoyance level.

Generally, there is an agreement that detectability and annoyance are strongly related to loudness and that a loudness model can potentially objectively predict and assess them. However, EVs operate in urban environments, so to account for the masking effect of background noise, a partial loudness model should be used. The model used for this paper was Moore-Glasberg's time-varying loudness, which computes loudness following an advanced model of the human ear based on the auditory filter bank concept.

Twenty-three people took part in three tests to obtain subjective detection thresholds. For the first test, subjects listened to warning sounds without background noise to familiarize them with the stimuli. The subjective reaction time was measured by asking the subject to press a button as soon as they heard a short noise impulse. The subject was then presented with one of the four warning sounds in the presence of one of five background noises, totaling 20 combinations. The second test used an adaptive force choice paradigm to reduce the effects of individual bias when detecting warning sounds in background noise. The subject was presented with three consecutive sound samples, all containing the same one-second segment of simulated urban noise, but one also containing a one-second segment of a warning sound.

The third test investigated the influence of background noise on perceived annoyance. The warning sounds were evaluated in five noise conditions and the subjects told to imagine themselves in an urban environment, for example, sitting outside a café.

The results confirm that the detection thresholds are heavily influenced by the subject’s confidence when they provide their responses. The detection thresholds in terms of partial loudness are similar for stationary warning sounds. The perceived annoyance increases with partial loudness as expected, and the model explains the effect of background noise on annoyance perception rather well.

Read the full white paper here