Bird Sound Classification using Deep Neural Networks: A Comparative Analysis of State-of-the-Art Models and Custom Architectures
Year: 2023
Type: article
Abstract: Bird sound classification plays a vital role in ecological monitoring and biodiversity conservation efforts. In this research paper, we explore the efficacy of Deep Neural Networks (DNNs) for this task, conducting a comparative analysis of five well-established methods: Xception, InceptionV3, ResNet50, EfficientNet, and VGG16. The BirdCLEF 2022 dataset, sourced from Xeno-Canto on Kaggle, serves as the foundation for our investigation. To extract essential acoustic features from the dataset, we employ Mel Frequency Cepstral Coefficients (MFCC). By converting the audio files into spectrograms, we enable the utilization of image-based classification techniques on this audio data. In addition to the state-of-the-art models, we design and implement two custom-made Convolutional Neural Network (CNN) architectures. These models surpass several existing approaches, achieving accuracy rates of 80.11% and 76.94%, respectively. Our research offers valuable insights into the performance and suitability of various DNN models for bird sound classification. Furthermore, the success of our custom architectures highlights the potential for tailored solutions in this domain. The outcomes of this study have implications for bird species identification, ecological monitoring, and wildlife conservation efforts, paving the way for further advancements in avian soundscape analysis.
Language: English
Topic: Music and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7217773356