South-Asian-Sounds_Audio-Classification

This repository contains the code for the research paper "South Asian Sounds: Audio Classification," published at the 4th International Conference on Computer, Communication, Control & Information Technology (C3IT) in 2024.

The project focuses on classifying diverse South Asian urban and cultural sounds by combining Mel-Frequency Cepstral Coefficients (MFCCs) with a custom 1D Convolutional Neural Network (1D-CNN).

---

🏗️ Model Architecture

Below is the architecture of the proposed SAS-CNN model used in this study.

---

✨ Key Highlights

• Audio recordings are preprocessed and segmented into standardized 4-second clips.
• Features are extracted using MFCCs to capture the unique spectral characteristics of each sound.
• Classification is performed using a custom 1D-CNN (SAS-CNN) and compared against baseline models like ResNet50V2.
• The model was evaluated on two key datasets: the newly introduced SAS-KIIT dataset and the benchmark UrbanSound8K dataset.
• High classification accuracy was achieved through 10-fold cross-validation:
  • 99.78% on the SAS-KIIT dataset.
  • 94.26% on the UrbanSound8K dataset.

---

🏆 Contribution

This research makes the following contributions:

• Proposes a robust and effective pipeline for classifying South Asian urban sounds.
• Introduces the SAS-KIIT dataset, a new public resource for the research community.
• Demonstrates that combining MFCC features with a 1D-CNN is a highly effective approach for multi-class sound recognition.

---

🔗 Links

• Published Paper (IEEE Xplore): https://ieeexplore.ieee.org/document/10829485
• SAS-KIIT Dataset: https://sas-kiit.netlify.app/

---

💻 Code Overview

This repository contains the Python scripts for the complete pipeline:

• Audio preprocessing and feature extraction (MFCCs).
• Implementation of the 1D-CNN (SAS-CNN) model.
• Scripts for model training, evaluation, and inference.

---

📄 Citation

If you find this work useful in your research, please cite the paper.

BibTeX

@INPROCEEDINGS{10829485,
  author={Chatterjee, Rajdeep and Bishwas, Pappu and Chakrabarty, Sudip and Bandyopadhyay, Tathagata},
  booktitle={2024 4th International Conference on Computer, Communication, Control & Information Technology (C3IT)}, 
  title={South Asian Sounds: Audio Classification}, 
  year={2024},
  volume={},
  number={},
  pages={1-6},
  keywords={Translation; Pollution; Smart cities; Biological system modeling; Surveillance; Noise; Urban planning; Feature extraction; Rail transportation; Mel frequency cepstral coefficient; Audio classification; CNN; MFCC; Sound recognition},
  doi={10.1109/C3IT60531.2024.10829485}
}