采用复倒谱峰值滤波GMM识别混响语音  

Using complex cepstrum peak filter for reverberation recognition by GMM

作　　者：孔荣[1] 吴迪[1] 廖启鹏[1] 朱俊杰[1] 周强[1] 陶智[1] 

机构地区：[1]苏州大学物理科学与技术学院,江苏苏州215006

出　　处：《计算机工程与应用》2014年第15期191-193,203,共4页Computer Engineering and Applications

摘　　要：针对混响环境下语音识别系统性能急剧下降问题,提出一种采用复倒谱峰值滤波GMM识别混响语音的方法。通过训练纯净语音的MFCC特征参数构建高斯混合模型,在识别混响语音前引入复倒谱峰值滤波器以减少混响引起的语音失真而提高混响环境下语音识别率。经实验验证,该方法避免了在现实条件下准确估计房间冲击响应函数的麻烦,降低了计算难度,提高了混响环境下至少4%的系统识别率。The performance of speech recognition system will fall sharply in reverberant environment. In order to solve it, this paper proposes a method to use complex cepstrum peak filter to improve reverberation speech recognition rates by GMM. The GMM is constructed by MFCC parameters of pure speech, before identifying, this paper introduces the complex cepstrum peak filter to decrease speech distortions and improve the recognition rate in reverberant environment. The experimental results show that this method avoids estimating the room impulse response function in real conditions accurately, reduces the computational difficulty, and improves more than 4%of the system recognition rates in reverberant environment.

关 键 词：高斯混合模型 复倒谱 Mel频率倒谱系数(MFCC) 

分 类 号：TN912.3[电子电信—通信与信息系统]