混响环境下语音信号优化识别仿真研究

摘要

对混响环境下语音信号进行优化识别,可以提高连续英汉语音的识别度.进行语音信号优化识别时,在混响环境下,应对连续英汉语音增强进行增强处理,但是采用当前的方法进行英汉语音增强时,需要大量的延时单元来进行语音增强,存在增强过程较为繁琐的问题.提出一种基于宽带波束形成的混响环境下语音信号优化识别方法.上述方法先融合于小波滤波器将原始的英汉语音变换到小波域,组建小波域阵列宽带英汉语音信号模型,并对双耳正前方英汉语音进行增强,并抑制其它方位噪声,利用改进的维纳滤波方法将各段波束输出英汉语音数据时间序列进行连接,组建连续的波束输出序列,实现对混响环境下连续英汉语音信号的优化识别.仿真结果证明,所提方法对语音信号优化识别较好,有效地抑制了混响和噪声对连续英汉语音信号的影响.%In this paper,we proposed an optimization recognition method of speech signal under the reverberation environment based on the wideband beamforming.Firstly,the initial English-Chinese speech was transformed into the wavelet domain integrated with the wavelet filter and the English-Chinese speech signal model of array broadband was built in the wavelet domain.Then,the English-Chinese speech ahead of ear was enhanced and the noise in the other position was suppressed.Moreover,the modified Wiener filtering method was used to connect the time series of English-Chinese speech data outputted by each beam.Finally,we built the continuous beam output series and achieved the optimization recognition of continuous English-Chinese speech signal under the reverberation environment.The simulation results show that the method has better recognition effect,which can suppress the influence of reverberation and noise on the continuous English-Chinese speech signal.