Road Condition Anomaly Detection using Self-Supervised Learning from Audio

This study proposes a method to extract driving signals from audio signals by eliminating non-driving noises through noise reduction techniques. The proposed methodology utilizes self-supervised learning techniques with a simple masking augmentation approach applied to the extracted driving signals. This approach enhances the accuracy of the model in identifying the driving signals by masking both time and frequency domains in the input data during training. The method's accuracy is evaluated using an internally collected dataset, demonstrating its effectiveness in producing accurate results. The proposed method has the potential to improve the accuracy of audio-based driving signal detection systems.