基于话者识别的新去噪和识别方法

摘要

In order to improve the robustness of the systems, second generation wavelet and niche and PSO are combined for speech denoising. In order to solve the identification rate of speaker recognition in dynamic environment, the arithmetic of niche and PSO is applied to GMM. The experiment ' s results are showed that the parameters proposed have produced better effect. Applying to GMM of niche and PSO further improve recognition performance.%为了提高系统的鲁棒性,联合二代小波和TEO一起对语音进行去噪.为了解决动态环境下的说话人识别的误识率问题,在识别阶段,把小生境粒子群算法应用于GMM之中.从实验得出,采用本文提出的参数对于说话人识别的效果较好.采用基于小生境粒子群的高斯混合模型进一步提高了识别性能.