基于PCANN/PDP混合结构的噪声环境下说话人识别方法的研究

提出了一种基于PCANN／PDP混合结构的实际环境下说话人识别的方法。它采用相继几帧组成的特征参数矢量作为说话人识别系统的输入，能有效地在说话人识别系统中引入帧间相关信息。针对噪声往往具有帧间相关性小，能量分布频率范围广且数值较小，在语音信号主分量特征中对应于贡献率较小的分量等特点，话人识别系统的前端增加语音参数压缩的主分量分析神经网络（PCANN）。同时提出了概率DP匹配说话人识别方法。通过对噪声环境下与文本无关的说话人识别实验，这种方法的有效性得到了验证。

Study on Speaker Recognition Under Noise Environment Based on PCANN/PDP

This paper presents the method for speaker recognition based on PCANN/PDP mixed configuration.It adopts a sequence of frames as input of speaker recognition system to introduce the correlation between frames.In allusion to the noise characteristics of small relativity,small contribution in main component of voice signal,it adds PCANN for compressing voice main parameters in the preceding part of speaker recognition system.At the same time,it advances speaker recognition of probability DP.Through the experiments of speaker-independent in noise environment,the validity of the proposed approach could be verified.