Sound Classification in Indian Cities Using Multi-Label Data and Transfer Learning

Rishi Gupta, Vaibhav Kumar

Keywords: Transfer learning, Indian Cities, Urban Audio Classification, Multi Label Classification

TL;DR: Complex Urban Sound Classification using Deep Learning

Abstract: The existing research primarily focuses on single-label classifications of complex urban sounds, ignoring crucial factors like sound mixtures and duration. This work proposes a novel transfer learning approach for multi-label sound classification. We fine-tuned a pre-trained VGGish model using a manually labeled audio dataset containing representative classes from diverse Indian cities, collected through various avenues. Our model achieves remarkable performance, demonstrating a significant 32% increase in F1-score compared to models trained on the AudioSet benchmark dataset.

Submission Number: 105