摘要
本文针对非平稳噪声和强背景噪声下声音信号难以提取的实际问题,提出了一种DCT域的维纳滤波方法。该方法进行了DCT域清浊音分割步骤,给出了DCT域频谱信噪比迭代更新机制与具体实施方案,设计了DCT域的二维维纳滤波。实验仿真表明,该算法能有效地去噪滤波,显著地提高语音识别系统性能与可懂度,且在不同的噪声环境和信噪比条件下具有鲁棒性。本算法计算代价小,简单易实现。
In this paper, a DCT domain Wiener filtering method is proposed due to the fact that voice signal is hard to be extracted when there is non-stationary noise and strong background noise. In the method the DCT domain voice segmentation step is made. The mechanism of DCT domain spectrum signal to noise ratio iteration is updated and specific implementation plan is given. The two dimensional Wiener DCT domain filtering is designed. Simulation experiments show that the algorithm can effectively de-noise, which significantly improves the speech recognition system performance and the intelligibility and reveals excellent robustness in different noise environments and SNR. The algorithm is low in calculation cost and is simple and easy to use.
出处
《湖南涉外经济学院学报》
2014年第1期83-88,共6页
Journal of Hunan International Economics University
关键词
语音增强
维纳滤波
离散余弦变换
自适应处理
speech enhancement
Wiener filtering
discrete cosine transformation
adaptive processing
