多重约束非负矩阵分解的非平稳噪声语音增强（英文）

发布时间：2018-04-14 20:45

本文选题：语音增强 + 低秩约束　；参考：《控制理论与应用》2017年06期

【摘要】：低信噪比非稳态噪声环境中的语音增强仍是一个开放且具有挑战性的任务.为了提高传统的基于非负矩阵分解(nonnegative matrix factorization,NMF)的语音增强算法性能,同时考虑到语音信号的时频稀疏特性和非稳态噪声信号的低秩特性,本文提出了一种基于多重约束的非负矩阵分解语音增强算法(multi-constraint nonnegative matrix factorization speech enhancement,MC NMFSE).在训练阶段,采用干净语音训练数据集和噪声训练数据集分别构建语音字典和噪声字典.在语音增强阶段,在非负矩阵分解目标函数中增加语音分量的稀疏性约束和噪声信号的低秩性约束条件,MC NMFSE能够更好地从带噪语音中获得语音分量的表示,从而提高语音增强效果.通过实验表明,在大量不同非平稳噪声条件和不同信噪比条件下,与传统的基于NMF的语音增强方法相比,MC NMFSE能获得较低的语音失真和更好的非稳态噪声抑制能力.
[Abstract]:Speech enhancement in unsteady noise environments with low SNR is still an open and challenging task.In order to improve the performance of the traditional speech enhancement algorithm based on nonnegative matrix factorization (NMF), the time-frequency sparse characteristic of speech signal and the low rank characteristic of unsteady noise signal are taken into account.In this paper, a multi-constraint nonnegative matrix factorization speech enhancement algorithm for speech enhancement based on multi-constraint nonnegative matrix decomposition is proposed.In the training stage, the clean speech training data set and the noise training data set are used to construct the speech dictionary and noise dictionary respectively.In the phase of speech enhancement, the sparse constraint of speech component and the low rank constraint condition of noise signal are added to the objective function of non-negative matrix decomposition. MC / NMFSE can better obtain the representation of speech component from noisy speech.In order to improve the speech enhancement effect.The experimental results show that under a large number of non-stationary noise conditions and different signal-to-noise ratio (SNR) conditions, compared with the traditional speech enhancement method based on NMF, MC / NMFSE can achieve lower speech distortion and better unsteady noise suppression ability.
【作者单位】：北京大学信息工程学院现代信号与数据处理实验室;
【基金】：Supported by National Natural Science Foundation of China(61271309) Shenzhen Science Research Program(CXZZ20140509093608290)
【分类号】：TN912.3

【相似文献】