Music Source Activity Detection and Separation Using Deep Attractor Network
Music Source Activity Detection and Separation Using Deep Attractor Network
复制标题
使用深度吸引子网络的音乐源活动检测和分离
DOI：
10.21437/interspeech.2018-2326
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
N. Mesgarani
中科院分区：
文献类型：
--
作者：
Rajath Kumar;Yi Luo;N. Mesgarani
在音乐信号处理中，歌声检测和音乐源分离是一个被广泛研究的课题。基于深度神经网络的源分离的最新进展已经推进了人声和乐器分离问题的性能状态，而联合源活动检测和分离的问题尚未探索。在本文中，我们提出了一种使用深度吸引子网络（DANet）在训练音乐源分离时生成的高维嵌入来执行源活动检测的方法。通过一起定义这两个任务，DANet能够根据活动源动态估计输出的数量。我们提出了一个期望最大化（EM）的训练范式DANet，进一步提高了原来的DANet的分离性能。实验表明，我们的网络实现了更高的源分离和可比的源活动检测对基线系统。
In music signal processing, singing voice detection and music source separation are widely researched topics. Recent progress in deep neural network based source separation has advanced the state of the performance in the problem of vocal and instrument separation, while the problem of joint source activity detection and separation remains unexplored. In this paper, we propose an approach to perform source activity detection using the high-dimensional embedding generated by Deep At-tractor Network (DANet) when trained for music source separation. By defining both tasks together, DANet is able to dynamically estimate the number of outputs depending on the active sources. We propose an Expectation-Maximization (EM) training paradigm for DANet which further improves the separation performance of the original DANet. Experiments show that our network achieves higher source separation and comparable source activity detection against a baseline system.
DOI：
10.23919/eusipco.2018.8553571
发表时间：
2018-03
期刊：
2018 26th European Signal Processing Conference (EUSIPCO)
影响因子：
--
作者：
Emad M. Grais;D. Ward;Mark D. Plumbley
通讯作者：
Emad M. Grais;D. Ward;Mark D. Plumbley