Evaluation of Environmental Sound Classification using Vision Transformer

Recently, attention-based vision transformers have achieved significant success in audio classification tasks. While most cases of vision transformers for audio classification focus on achieving satisfactory scores in the end, there is still a lack of detailed evaluation from a practical standpoint. In this study, we conducted a comparative study, boosting vision transformers on the ESC-50 dataset step by step. Our goal is to provide practitioners with a solid foundation for adapting vision transformers to the general Environmental Sound Classification task. Our comparative study encompasses various aspects, including model setting, data augmentation, cross-domain transfer learning, and model pruning, offering practical insights for the implementation of vision transformers in real-world scenarios.