范数正则化解相关集成学习基音频率检测

发布时间：2018-03-29 21:05

本文选题：低信噪比环境　切入点：基音频率　出处：《计算机工程与应用》2017年11期

【摘要】：低信噪比环境下的基音频率检测极其重要且富有挑战性,至今未得到很好的解决。基于此,首先构造了基于PEFAC的频域空间检测模型,将基音频率作为特征进行提取,然后提出范数正则化的解相关集成学习神经网络模型(L2-DNNE)对其进行训练,利用负相关学习机制(NCL)和模型复杂度约束项提高集成学习模型的泛化能力,从而获取基音频率的最优值,且在测试精度和时间代价上取得了较好的平衡。将该算法与相关有代表性的算法进行比较。比较结果表明,该算法在不同类型不同程度的噪声环境下,能显著提升检测识别率,尤其在低信噪比下有更显著优势。
[Abstract]:Pitch frequency detection in low signal-to-noise ratio (SNR) environment is very important and challenging, which has not been solved well. Based on this, a frequency domain spatial detection model based on PEFAC is constructed, and pitch frequency is extracted as a feature. Then the norm regularization model of decorrelation ensemble learning neural network (L2-DNNEs) is proposed to train it. The negative correlation learning mechanism (NCL) and the complexity constraints of the model are used to improve the generalization ability of the integrated learning model, so as to obtain the optimal pitch frequency. This algorithm is compared with the related representative algorithms. The comparison results show that the algorithm can significantly improve the detection and recognition rate in different types of noise environments with different degrees of noise. Especially in the low signal-to-noise ratio has a more significant advantage.
【作者单位】：重庆广播电视大学;重庆大学信号与信息处理研究所;
【基金】：国家自然科学基金(No.61108086,No.91438104,No.61571069,No.81601970,No.61501065) 中国博士后科学基金(No.2013M532153) 中央高校基本科研业务费专项资金(No.CDJZR155507) 重庆市教委科学技术研究项目(No.KJ1603805) 重庆市基础科学与前沿技术研究专项(No.cstc2016jcyj A0043,No.cstc2016jcyj A0064,No.cstc2016jcyj A0134) 重庆市社会事业与民生保障科技创新专项(No.cstc2016shmszx40002) 重庆市博士后科研项目特别资助教育部留学回国人员基金
【分类号】：TN912.3;TP183

【相似文献】