摘要
麦克风采集的语音信号常含有噪声和混响,会降低语音信号的清晰度和可懂度,导致语音识别系统识别率降低。对Habets双麦克风噪声和混响消除算法研究发现,算法具有很好的噪声消除性能,但是输出语音的识别率较低。在其基础上提出一种改进算法,通过引入维纳滤波器,对后置OM-LSA算法进行替换。经验证,改进算法的输出失真减小,识别率提升。
Voice signal captured by mic usually has noise and reverberation which will decrease the articulation and intelligibility of the signal and lead to the recognition rate degradation of the voice system. Based on the research of Habets dual mic noise and reverberation elimination algorithm, it has excellent performance on noise elimination but low recognition rate with output voice. This article introduces an improved version based on the algorithm by using Wiener filter to replace the post OM-LSA algorithm. Test proves that the output distortion of improved algorithm decreases and its recognition rate raises.
出处
《信息技术与网络安全》
2018年第2期127-130,共4页
Information Technology and Network Security
