稀疏卷积非负矩阵分解的语音增强算法

发布时间：2018-11-17 16:25

【摘要】：鉴于卷积非负矩阵分解在语音增强算法中的成功应用,进一步考虑语音信号在时频域的稀疏性,提出了稀疏卷积非负矩阵分解(Sparse convolutive nonnegative matrix factorization,SCNMF)的语音增强算法。该算法包括训练和增强两个阶段。训练阶段通过SCNMF算法分别对纯净语音和噪声的频谱进行训练,得到纯净语音和噪声字典,并将其作为增强阶段的先验信息。增强阶段首先通过SCNMF算法对带噪语音的频谱进行分解,然后利用纯净语音和噪声联合字典以及相应的迭代公式对语音编码矩阵进行估计,重构增强语音。通过实验仿真分析了稀疏因子对增强语音质量的影响。实验结果表明,在非平稳噪声和低信噪比条件下,本文算法增强效果均优于多带谱减、非负矩阵分解和卷积非负矩阵分解等传统的算法。
[Abstract]:In view of the successful application of convolutional nonnegative matrix factorization in speech enhancement algorithm, a speech enhancement algorithm based on sparse convolutional nonnegative matrix factorization (Sparse convolutive nonnegative matrix factorization,SCNMF) is proposed, considering the sparsity of speech signal in time-frequency domain. The algorithm includes two stages: training and enhancement. In the training stage, the pure speech and noise dictionaries are obtained by training the spectrum of pure speech and noise by SCNMF algorithm respectively, and they are regarded as the prior information of the enhancement stage. In the enhancement phase, the spectrum of noisy speech is decomposed by SCNMF algorithm, and then the speech coding matrix is estimated by using the joint dictionary of pure speech and noise and the corresponding iterative formula to reconstruct the enhanced speech. The effect of sparse factor on speech quality enhancement is analyzed by simulation. The experimental results show that the enhancement effect of this algorithm is better than that of traditional algorithms such as multi-band spectral subtraction, non-negative matrix decomposition and convolution non-negative matrix factorization under the condition of non-stationary noise and low signal-to-noise ratio.
【作者单位】：解放军理工大学指挥信息系统学院;西安通信学院基础部;
【基金】：江苏省自然科学基金(BK2012510)资助项目
【分类号】：TN912.3

【参考文献】