论文部分内容阅读
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, aiming at the practical problem of difficult to extract the sound signal under non-stationary noise and strong background noise, a Wiener filtering method in DCT domain is proposed. In this method, the DCT domain is divided into voiced and unvoiced segments, and the iterative update mechanism and specific implementation of spectral signal-to-noise ratio in DCT domain are given. The two-dimensional Wiener filter in DCT domain is designed. The experimental results show that this algorithm can effectively denoise the filter and improve the performance and intelligibility of the speech recognition system significantly, and it is robust under different noise environments and SNR. The algorithm is less costly and simple to implement.