@inproceedings{7e7d5c2004964dc6bfd837db20fb92c4,
title = "Optimization of EVS speech/music classifier based on deep learning",
abstract = "EVS (Enhanced Voice Services) is a multi-mode codec proposed by 3GPP (3rd Generation Partnership Project) for 4G mobile services with a good performance and codec quality. The key technology of EVS lies in the flexible switch between speech and audio coding mode which mostly depends on the speech/music classifier. In general, the music signal is more complex than speech signal, and it conform less to any known LP (Linear Prediction)-based model. Taking the EVS's internal classifier as a baseline system, this study presents the optimization of the speech/music classifier from the perspective of neural network. The paper demonstrates the effectiveness of the optimized system on the MUSAN database. The experimental results show that the optimized system can improve the performance of the classifier, especially for music classification. Performed subjective experiments indicate that the proposed classification architecture improves perceived audio quality of the EVS codec.",
keywords = "Audio test, Deep Learning, EVS, Speech/Music classifier",
author = "Zhitong Li and Xiang Xie and Jing Wang and Volodya Grancharov and Wei Liu",
note = "Publisher Copyright: {\textcopyright} 2018 IEEE.; 14th IEEE International Conference on Signal Processing, ICSP 2018 ; Conference date: 12-08-2018 Through 16-08-2018",
year = "2019",
month = feb,
day = "2",
doi = "10.1109/ICSP.2018.8652295",
language = "English",
series = "International Conference on Signal Processing Proceedings, ICSP",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
pages = "260--264",
editor = "Yuan Baozong and Ruan Qiuqi and Zhao Yao and An Gaoyun",
booktitle = "2018 14th IEEE International Conference on Signal Processing Proceedings, ICSP 2018",
address = "United States",
}