Abstract: Acoustic signal is difficult to be extracted under noise with strong background non-stationary noise, a Wiener filtering method is proposed at DCT domain. The paper details segmentation steps between the voiceless and voiced, to give the SNR iterative update method and the concrete implementation plan on DCT domain spectrum, to design the two-dimension Wiener filtering on DCT domain. Experimental simulation shows that the algorithm can effectively denoise and filter, improve the intelligibility, and is robust in different noise environment and SNR conditions. The algorithm computational cost is small, easy to realize.

Key words: speech enhancement, Wiener filtering, discrete cosine transform, adaptive processing

摘要： 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。列出了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，改善可懂度，且在不同的噪声环境和信噪比条件下具有鲁棒性。该算法计算代价小，简单易实现。

关键词: 语音增强, 维纳滤波, 离散余弦变换, 自适应处理