Aplication de modéles de réseaux de neuronnes convoolutifs pré entraines a la classification des différent types de bruits sonores environnementaux
No Thumbnail Available
Date
2026
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Several approaches to environmental sound classification have been proposed to address the main challenges of automatic audio signal recognition. Particular attention has been given to deep learning and transfer learning, especially convolutional neural networks (CNNs), which have proven to be effective solutions in this field.
In this work, we present a comparative study of two pre-trained architectures, ConvNeXtBase and EfficientNetV2S, applied to the classification of environmental sounds from the ESC-50 dataset. We selected 26 sound classes and used two visual representations of the audio signal, namely spectrograms and Mel-spectrograms, in order to evaluate their impact on classification performance. Both models were trained and optimized in the Google Colab environment, taking advantage of its available GPU resources. Based on the results, we observe that the configuration combining EfficientNetV2S with the Mel-spectrogram achieved the best accuracy, with a score of 97.88%.