弱监督音频事件检测算法的一种新型神经网络结构
首发时间:2020-04-08
摘要:2017年谷歌推出大规模弱监督音频数据集Audio Set后,使用弱监督数据研究音频事件检测算法成为音频事件检测领域的主流。但是,目前弱监督音频事件检测算法存在性能差的问题。为提高弱监督音频事件检测算法的性能,本文基于该领域任务特点,提出了一种新型神经网络结构SEDNet。在DCASE 2017 task4数据集上,子任务音频事件分类的F1值为0.579,和基准Baseline CRNN相比提高了6.34%,子任务音频事件定位的Error Rate为0.562,相比Gated CRNN降低了3.38%。弱监督音频事件检测算法性能的提高有助于推动算法的落地。
关键词: 人工智能 聚合函数 多示例学习 弱监督音频事件检测
For information in English, please click here
A New Neural Network for Weakly Supervised Sound Event Detection
Abstract:Google released the large-scale weakly supervised dataset Audio Set in 2017. Since then, researchers around the world use it and its subsets to train the sound event detection system. However, weakly supervised sound event detection system does not perform well. In this paper, a new neural network SEDNet is proposed to improve the performance of weakly supervised sound event detection system. The proposed method is evaluated on Task4 of Detection and Classification of Acoustic Scenes and Events (DCASE) 2017 Challenge in conjunction with other systems and shows a better performance. The F1 score of sound event classification is 0.579, which is 6.34% higher than Baseline CRNN, and the error rate of sound event location is 0.562, which is 3.38% lower than Gated CRNN.
Keywords: Artificial Intelligence Pooling Function Multiple Instance Learning Weakly Supervised Sound Event Detection
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
弱监督音频事件检测算法的一种新型神经网络结构
评论
全部评论0/1000