|
Dear Frederico,
I just want to contribute with a little field experience although I have no references relevant to your classification problem.
Before 2020 I have published works and a patent on the classification of asphalt roughness and wetness from microphones placed inside a car. The interference of the engine exhaust was low and it was notable only when the mike (omnidirectional) was placed at
20-40cm from the exhaust pipe, therefore the engine sound is nowadays hard to use as a classification hint for deep learning algorithms. In recent cars (let's say Euro5 onward) the engine acoustic emissions are hardly heard at speeds >50 km/h, cause wind and
tyre-road noise are predominant. On highways, the presence of a 6th gear or automatic will also make the engine spin at relatively low RPM, making the task more difficult, while at low speeds your problem may be easier to tackle. Take care
of producing a dataset that involves different floor conditions and roughness, cause these will greatly affect the sound produced by passing vehicles and a good generalization requires involving several tarmac types.
Do consider possible biases: it may be easy to discriminate older vehicles (with higher engine acoustic emissions and/or with high RPM on highways) from recent ones, but more difficult to discriminate EVs vs. internal combustion vehicles. Bear
this in mind when preparing your dataset and evaluation because it may bias the result, or conversely, removing relatively silent IC cars from the dataset may help you gain nice results but bad real-world performance. Finally, I see source localization as
a major problem if you don't use very directional microphones or beamforming, cause many cars will be passing by at toll plazas or highways.
For some acoustic information on different terrain types you can give a look at the seminal paper from Abdic et al. related to wetness detection: "Detecting Road Surface Wetness from Audio: A Deep Learning Approach".
For some more info on the effect of roughness take a look at the papers: "Deep Neural Networks for Road Surface Roughness Classification from Acoustic Signals" (144 AES convention), "Processing Acoustic Data
with Siamese Neural Networks for Enhanced Road Roughness Classification"(IJCNN2019)
Good luck with your research work
From: AUDITORY - Research in Auditory Perception <AUDITORY@xxxxxxxxxxxxxxx> on behalf of Frederico Pereira <pereira.frederico@xxxxxxxxx>
Sent: Tuesday, January 2, 2024 7:54 PM To: AUDITORY@xxxxxxxxxxxxxxx <AUDITORY@xxxxxxxxxxxxxxx> Subject: [AUDITORY] Electric vehicles acoustic discrimination Dear Auditory List,
I hope this message finds you well.
I am currently in search of published research pertaining to the discrimination of electric vehicles compared to internal combustion vehicles based on noise emissions. Specifically, I am interested in studies that explore the feasibility of implementing
such discrimination systems in highway environments (such as installed in span structures, toll plazas, etc..).
I acknowledge inherent challenges associated with this task, including issues related to extraneous noise, the utilization of Acoustic Vehicle Alerting Systems (AVAS), and, notably(?), the dominance of tire-road noise at higher velocities. My particular
interest lies in uncovering research that delves into specific acoustic features that may be extracted from microphone array beamforming systems.
From searches on Google Scholar, I was unable to find substantial literature on this specific topic. Therefore, I am reaching out to the esteemed members of this community to inquire if any of you could direct me towards relevant literature addressing
this topic.
Thank you in advance for your assistance.
- Frederico
--
Frederico Pereira
Mobile:+351 937356301 Audio and Acoustics, Perception Interaction and Usability
|