8–12 Sept 2026
Europe/Vienna timezone

Evaluation of Psychoacoustic Tonality Models for Cabin Noise Across Vehicle Powertrains

FA2026/797
8 Sept 2026, 16:20
20m
Saal 4 (Messe Congress Graz)

Saal 4

Messe Congress Graz

A19 Product Sound Quality and Sound-Driven Design A19.02 Listening experience in sound-driven design

Speaker

Zhenxian Li (INSA Lyon, LVA UR677)

Description

Tonality strongly affects the perceived quality and annoyance of vehicle interior noise. This study evaluates two widely used tonality metrics—the Aures tonality metric and the Sottek psychoacoustic Hearing Model tonality metric (ECMA-418-2)—against subjective ratings collected in two complementary listening experiments. Stimuli were drawn from chassis-dynamometer recordings of four vehicles representing different propulsion technologies (two electric, one hybrid-electric, one diesel), captured at the driver and rear-passenger head positions under full and partial throttle. Study 1 used 112 unprocessed 3 s segments rated by 33 listeners. Study 2 used 48 order-tracking resynthesised stimuli, which preserved the spectral and tonal structure while suppressing temporal fluctuations and equalising loudness across stimuli; each of 19 listeners rated every stimulus three times in randomised order. Hierarchical clustering of Study 1 ratings revealed two listener subgroups with distinct evaluation strategies: one showed a moderate association with the Sottek metric (r = 0.59) but none with Aures, whereas the other showed only weak associations with either metric; ratings in both clusters also correlated with loudness. In Study 2, with loudness and temporal fluctuations controlled, both tonality metrics were strong predictors of subjective ratings (Sottek: r = 0.77; Aures: r = 0.72). Overall, the Sottek metric outperforms the Aures metric for complex, realistic stimuli, while both metrics perform comparably well under controlled conditions. Although tonality alone is a difficult perceptual attribute for non-expert listeners to evaluate, both metrics remain valuable as input features for higher-level perceptual models such as annoyance prediction.

Authors

Zhenxian Li (INSA Lyon, LVA UR677) Etienne Parizet (INSA Lyon, LVA UR677)

Presentation materials