![]() |
Proceedings of the 11th Convention of the European Acoustics Association Forum Acusticum / EuroNoise 2025 Málaga, Spain June 23 - 26, 2025 |
|
Abstract Long-term noise monitoring is essential to ensure compliance with regulations. This process requires the removal of spurious sounds unrelated to the target source or to the typical soundscape of the monitored area. Traditionally, such tasks relied on manual labelling by operators, but recent advancements in data-driven methodologies highlight that it is time to automate the process using cutting-edge machine learning techniques. Pre-trained models, widely available in literature, are trained on extensive datasets covering numerous classes and serve as a foundation for developing specialized machine learning models fine-tuned for specific tasks or subsets of classes. This study presents a Transfer Learning approach to leverage the knowledge of the Contrastive Language-Audio Pretraining (CLAP) model for a classification task focused on a subset of its original classes. Although the CLAP model has demonstrated adaptability across a broad range of classes with good results, the findings of this study suggest that the application of Transfer Learning can enhance classification accuracy for the selected subset of environmental sound classes. |
||||||||||||||||||||