Communication Dans Un Congrès Année : 2025

Automatic detection of roadway noise using AI classifiers and sound level features

Détection automatique de bruit routier à l'aide de modèles de classification automatique et d'analyse de l'évolution du niveau sonore

Abbes Kacem
  • Fonction : Auteur
  • PersonId : 1185612
  • IdRef : 243136641
Nadine Martin

Résumé

Noise management has become a major public health issue over several decades. To initiate protective measures against noise overexposure, it is essential to accurately evaluate annoyance. This involves detecting sound sources, their emission durations and associated sound levels. Among commonly encountered sources, road traffic is a prominent contributor to noise pollution, according to health organisation reports. This paper presents an automatic roadway noise detection system. The proposed method combines event detection and classification through a multi-layer approach. Sound event detection is ensured by distinct units. First, a sliding window enables signal preclassification by identifying specific patterns on its mel-spectrogram. Simultaneously, sound level features are used to detect prominent periods, often marking off sound events. Both units combined and sharpened provide detection of the most relevant sound events over the acoustic signal. Precise classification is issued by an additional AI model, dedicated to recognising roadway noise among various vehicle types. Our system operates in diverse soundscapes while maintaining a high level of roadway noise detection accuracy. It permits an automatic estimation of roadway noise contribution, which corresponds to equivalent sound level when roadway noise prevails. This estimation closely aligns with manual assessments from experts, validating the proposed system relevance.

Fichier principal
Vignette du fichier
Automatic_detection_of_Roadway_Noise_using_AI_classifiers_and_signal_processing_algorithms.pdf (799.17 Ko) Télécharger le fichier
Origine Fichiers éditeurs autorisés sur une archive ouverte
Licence

Dates et versions

hal-05268793 , version 1 (19-09-2025)

Licence

Identifiants

  • HAL Id : hal-05268793 , version 1

Citer

Erwann Betton-Ployon, Abbes Kacem, Jérôme Mars, Nadine Martin. Automatic detection of roadway noise using AI classifiers and sound level features. FA 2025 - Forum Acusticum - Euronoise, Jun 2025, Màlaga, Spain. ⟨hal-05268793⟩
99 Consultations
101 Téléchargements

Partager

  • More