| Title |
Boosting Principal Frequency based Data Augmentation for Sound Event Detection |
| Authors |
남지원(Jiwon Nam) ; 박상욱(Sangwook Park) |
| DOI |
https://doi.org/10.5573/ieie.2024.61.7.77 |
| Keywords |
Boosting principal frequency; Dcase; Sound event detection; Data augmentation; Audio&speech |
| Abstract |
Sound event detection aims to identify sounds of interest into the type of sounds and each interval. So, it is a crucial technique of audio-based situation awareness, multimedia analysis, and monitoring systems. To improve detection performance, an acoustic model that reflects the statistic of target sounds, which occur in various situations, is needed. However, it is difficult to consider various situations due to high cost of audio collection and preparing training dataset. To address this challenge, this paper analyzes the effectiveness of conventional data augmentation approaches applicable to training acoustic models efficiently, and then proposes a new data augmentation method. In the experiments based on the testbed provided by DCASE2020 task4, the proposed method shows the best performance in an event-based F1 score of 39.39% compared to other comparison groups such as time shiting, time rolling, masking, filtering, adding noise, and mix-up. |