South Asian Sounds: Audio Classification
Rajdeep Chatterjee, Pappu Bishwas, Sudip Chakrabarty, Tathagata Bandyopadhyay · 2024
Sound classification is a significant task in the field of audio processing, with applications ranging from urban planning to noise pollution monitoring. We use Mel-Frequency Cepstral Coefficients (MFCCs) and a 1D Convolutional Neural Network (1D-CNN) model to try to solve the problem of putting urban sounds from Bangladesh into different groups. The proposed approach involves extracting MFCC features from audio recordings of urban environments in Bangladesh, which capture the spectral characteristics of the sounds. A 1D-CNN model then processes these MFCC features to classify the sounds into predefined categories like road-traffic noise, traditional Bengali songs, trains (railways), classroom noise, tanpura, etc. We also assess the performance of our approach on a dataset that has been collected and recorded from various locations in India and Bangladesh, as well as the well-known UrbanSounds8k dataset. Experimental results demonstrate the effectiveness of the proposed method, achieving high classification accuracy in distinguishing between different urban sound classes across both datasets. Our findings suggest that the combination of MFCCs and 1D-CNNs offers a robust solution for urban sound classification, with potential applications in urban planning, environmental monitoring, and smart city initiatives.