| 何冲,王冬霞,王旭东,蒋茂松.一种基于正交非负矩阵分解的多通道线性预测语音去混响方法[J].声学技术,2018,37(5):465~474 |
| 一种基于正交非负矩阵分解的多通道线性预测语音去混响方法 |
| Speech dereverbration based on MCLP using orthogonal NMF |
| 投稿时间：2017-06-29 修订日期：2017-08-13 |
| DOI：10.16300/j.cnki.1000-3630.2018.05.011 |
| 中文关键词: 麦克风阵列 去混响 多通道线性预测 非负矩阵分解 |
| 英文关键词: microphone array dereverbration Multi-Channel Linear Prediction (MCLP) Non-negative Matrix Factorization (NMF) |
| 基金项目:辽宁省自然科学基金（201302022）资助项目。 |
|
| 摘要点击次数: 1164 |
| 全文下载次数: 976 |
| 中文摘要: |
| 在相对封闭的声学环境中，由于受到混响的影响，麦克风阵列采集到的信号清晰度降低、甚至混淆不清。为了解决这一问题，文章在多通道线性预测（Multi-Channel Linear Prediction，MCLP）语音去混响的基础上，提出了一种改进的多通道线性预测（Multi-Channel Linear Prediction，MCLP）方法即正交非负矩阵线性预测（Orthogonal Non-negative Matrix Factorization Multi-Channel Linear Prediction，ONMFMCLP）方法。该方法利用纯净语音的短时谱域的稀疏性，构建了基于正交的非负矩阵分解（Non-negative Matrix Factorization， NMF）的 Kullback-Leibler（KL）问题，通过对矩阵求迹、利用梯度下降法给出迭代规则，进而改进了MCLP中目标信号矩阵的协方差估计。实验结果表明，相对于其他方法，ONMFMCLP方法具有更好的去混响效果。 |
| 英文摘要: |
| In a relatively closed acoustic environment, the speech signals can be severely affected by reverberation, which degrades the intelligibility of speech and even results in confusion. In order to solve this problem, this paper presents a new dereverbration algorithm called ONMFMCLP based on the well-known multi-channel linear prediction (MCLP). This algorithm utilizes the sparse nature of clean speech in the short time spectrum domain to construct the KL optimization problem based on the orthogonal NMF. The iterative rules are given through solving the matrix trace and utilizing the gradient descent method, thereby improving the signal covariance matrix in the MCLP algorithm. Experimental results show that the ONMFMCLP algorithm can achieve a better dereverberation performance compared with other algorithms. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
