Electric Vehicle Cabin Noise: A New Model for Measuring Annoyance
As electric vehicles (EVs) continue to gain traction in the global automotive market, automakers are shifting their focus beyond range and performance to a more nuanced aspect of the driving experience—acoustic comfort. With the absence of engine noise, the cabin environment in EVs has become significantly quieter, but this silence has also brought new challenges. Subtle sounds such as wind noise, road noise, and the high-pitched whine of electric motors, once masked by combustion engines, are now more perceptible. These noises, though lower in overall sound pressure, can have a disproportionate impact on driver and passenger comfort. Understanding and quantifying how these sounds affect human perception is critical for enhancing the overall quality of the EV experience.
A recent study published in Noise and Vibration Control sheds new light on this issue by introducing a refined model for evaluating the psychoacoustic annoyance of interior noise in electric vehicles. Led by Wang Weidong from Pan Asia Technical Automotive Center Co., Ltd., in collaboration with Miao Zhenjing and Huang Yu from the State Key Laboratory of Mechanical System and Vibration at Shanghai Jiao Tong University, the research presents a data-driven approach to measuring how cabin noise influences subjective annoyance under real-world driving conditions.
Unlike many previous studies that relied on controlled laboratory environments or simulated noise recordings, this research takes a more practical and holistic approach. The team conducted on-road tests using a production electric vehicle, capturing cabin noise across four distinct road surfaces—smooth asphalt, concrete, gravel, and uneven bumpy terrain—at varying speeds ranging from 10 km/h to 110 km/h. This methodology ensures that the findings reflect actual driving scenarios, incorporating the complex interplay of wind, road texture, and mechanical noise that defines the EV auditory experience.
The researchers recorded audio at multiple seating positions inside the cabin, ultimately selecting the left ear position of the front passenger for subjective evaluation. A total of 16 audio samples were extracted, each representing a unique combination of speed and road condition. To ensure accuracy in subjective testing, the team meticulously calibrated the playback levels using a high-fidelity artificial head system, matching the reproduced sound pressure levels to the original on-road measurements. This attention to detail is crucial, as even minor discrepancies in volume can skew human perception and compromise the validity of subjective assessments.
To gauge how these noise samples affected human listeners, the team conducted a formal subjective evaluation with 30 participants—15 male and 15 female, aged between 21 and 29, a demographic that aligns closely with the core EV consumer base. None of the participants had a history of hearing impairment or psychological conditions that could influence their perception. The evaluation employed a modified version of the ICBEN (International Commission on the Biological Effects of Noise) annoyance scale, which ranges from 0 to 10, with additional semantic descriptors in Chinese to enhance clarity and consistency in responses. The scale was further refined by expanding the range of each rating category, allowing participants to express more granular differences in their annoyance levels.
Each participant listened to the 16 noise samples in a randomized order to eliminate sequence bias. After each 5-second playback, they provided a numerical rating based on how annoying they found the sound. The results revealed a clear trend: as vehicle speed and road roughness increased, so did the reported annoyance. For instance, driving at 50 km/h on a smooth asphalt road produced a relatively low annoyance score, while the same vehicle traveling at 60 km/h on a bumpy surface elicited a significantly higher response. The subjective ratings correlated strongly with A-weighted sound pressure levels, confirming that overall loudness remains a dominant factor in human perception of noise.
However, the study goes beyond simple loudness metrics by analyzing a suite of psychoacoustic parameters—objective measures that attempt to quantify subjective auditory experiences. These include loudness (measured in sone), sharpness (acum), roughness (asper), fluctuation strength (vacil), and tone-to-noise ratio (tonality, tu). Loudness, calculated according to the ISO 532-1 standard, reflects the perceived intensity of sound. Sharpness, based on the Aures model, captures the proportion of high-frequency energy in a signal—higher values indicate a more piercing or “metallic” sound, often associated with electric motor whine. Roughness and fluctuation strength describe the temporal variations in amplitude, with roughness linked to rapid amplitude modulations (typically above 20 Hz) and fluctuation strength to slower modulations (around 4–20 Hz), often perceived as a pulsing or throbbing sensation.
The analysis of these parameters revealed several key insights. As expected, loudness increased with both speed and road roughness. At higher speeds, aerodynamic noise and tire-road interaction generate more energy across the frequency spectrum. On rougher surfaces, the suspension and chassis transmit more vibrations into the cabin, contributing to a denser, more intrusive sound field. Sharpness also increased with speed, a finding that aligns with the physics of electric motors: as rotational speed rises, so do the fundamental and harmonic frequencies of electromagnetic forces acting on the motor structure, pushing more acoustic energy into the upper mid and high frequencies where human hearing is most sensitive.
What makes this study particularly valuable is its exploration of how these psychoacoustic parameters interact to influence overall annoyance. While loudness was the strongest predictor, the researchers found that sharpness, roughness, and fluctuation strength also played significant roles. Through statistical analysis, they discovered that when the influence of loudness was controlled, sharpness emerged as a key secondary factor. This suggests that even at similar overall volume levels, a noisier sound with more high-frequency content is perceived as more annoying—a critical consideration for EV designers aiming to fine-tune motor and inverter characteristics.
Roughness showed a complex relationship with annoyance. Before controlling for loudness, it appeared positively correlated with annoyance—rougher sounds were more irritating. However, after accounting for loudness, the correlation reversed, becoming negative. This counterintuitive result may indicate that in real-world driving, roughness often co-occurs with high loudness, but when isolated, moderate levels of amplitude modulation might not be inherently unpleasant, or could even provide a sense of dynamic feedback that some drivers associate with performance.
Fluctuation strength, which relates to low-frequency pulsations, also gained significance after controlling for loudness. This parameter is often linked to powertrain dynamics, such as torque ripple in electric motors or the cyclic nature of regenerative braking. Its increased relevance suggests that rhythmic, low-frequency variations in cabin noise—though subtle—can contribute to long-term fatigue and discomfort, especially during extended drives.
One of the most notable findings of the study is the relatively minor role played by tonality—the prominence of pure tones within the noise spectrum. In many laboratory-based studies, tonality has been shown to significantly increase annoyance, particularly when a distinct whine or hum is present. However, in this real-world dataset, tonality did not show a statistically significant correlation with subjective annoyance. The researchers hypothesize that in actual driving conditions, broadband noise from wind and road surfaces may mask pure tones, reducing their perceptual impact. This insight challenges the assumption that tonality reduction should always be a top priority in EV sound design and suggests that efforts might be better spent on managing overall loudness and sharpness.
Building on these findings, the team evaluated three existing psychoacoustic annoyance models to determine which best predicted the subjective responses. The first, developed by Fastl and Zwicker, combines loudness, sharpness, roughness, and fluctuation strength in a non-linear formula. The second, proposed by Di et al., adds tonality as an additional factor. The third, introduced by More, also includes tonality but uses a more complex, non-linear weighting scheme derived from subjective data.
The results were striking. The original Zwicker model outperformed both modified versions, achieving a coefficient of determination (R²) of 0.905 and the lowest mean squared error. This means it explained over 90% of the variance in subjective annoyance ratings, a remarkably high level of accuracy for a psychoacoustic model. In contrast, the models incorporating tonality performed significantly worse, with R² values of 0.834 and 0.568, respectively.
The superior performance of the Zwicker model, despite its lack of a tonality term, reinforces the idea that in real-world EV operation, the masking effect of broadband noise diminishes the impact of pure tones. It also suggests that the tonality weighting factors used in the other models—developed for aircraft and transformer noise—may not be applicable to the unique acoustic signature of electric vehicles. The Zwicker model, originally designed for general transportation and HVAC noise, appears to be more universally robust.
This finding has important implications for the automotive industry. Rather than adopting overly complex models with numerous parameters, engineers may achieve more accurate and reliable predictions using a well-established, simpler framework. The Zwicker model’s strong performance also validates its use as a benchmark for future EV sound quality assessments, potentially streamlining the development process and reducing reliance on extensive subjective testing.
The study’s methodology sets a new standard for psychoacoustic research in automotive applications. By combining real-world data collection, rigorous calibration, controlled subjective evaluation, and advanced statistical modeling, the authors have created a comprehensive and reproducible framework. Their approach bridges the gap between theoretical psychoacoustics and practical vehicle development, offering a tool that can be directly applied in NVH (noise, vibration, and harshness) engineering.
For automakers, the implications are clear. As EVs become increasingly common, the battle for market share will be fought not just on range and charging speed, but on the subtleties of the driving experience. A vehicle that feels quiet and refined, even at high speeds or on rough roads, will stand out in a crowded market. This research provides a scientific basis for optimizing cabin acoustics, guiding decisions on everything from motor design and gear selection to tire choice and aerodynamic shaping.
Moreover, the findings underscore the importance of holistic sound management. While reducing motor whine is important, it may not be the most effective strategy if the overall loudness and sharpness of the cabin environment are not addressed. Engineers should focus on a balanced approach—damping structural vibrations, improving sound insulation, and tuning the frequency response of the interior space to minimize high-frequency content.
The study also opens new avenues for future research. For example, how do these findings apply to different vehicle classes—compact hatchbacks versus luxury sedans? How does cabin noise annoyance vary with driving mode, such as sport versus eco? And how do individual differences—such as age, hearing sensitivity, or cultural background—affect perception? While the current study focused on a young, urban demographic, broader studies could reveal additional layers of complexity.
In conclusion, the work by Wang Weidong, Miao Zhenjing, and Huang Yu represents a significant advancement in the field of automotive psychoacoustics. By grounding their analysis in real-world data and validating their models against human perception, they have provided a reliable tool for evaluating and improving the acoustic comfort of electric vehicles. Their findings not only enhance our understanding of how sound affects human experience but also offer practical guidance for engineers striving to create quieter, more pleasant cabin environments.
As the automotive industry continues its transition to electrification, studies like this will play a crucial role in shaping the future of mobility. The goal is not just to build cars that are sustainable and efficient, but ones that are truly enjoyable to drive. And sometimes, the difference between a good EV and a great one comes down to what you don’t hear.
Wang Weidong, Miao Zhenjing, Huang Yu, Noise and Vibration Control, DOI: 10.3969/j.issn.1006-1355.2024.03.026