基于改进谱减法的语音增强

针对传统谱减法具有残余音乐噪声过强，清音部分损失严重的缺点，提出了一种利用语音信号的短时平均幅度差特征（AMDF）并结合短时平均幅度（AM）的语音检测算法，在原始语音估计式中引入了参数α和β，对传统的谱减法进行改进。根据采集的真实带噪语音数据，将传统谱减法和改进的谱减法结果进行了比较分析。在研究基于改进的谱减法的语音增强算法基础上，构建了以TMS320VC5509和TLV320AIC23 Codec为核心器件的实时系统，能够有效地提高语音信噪比．信号的信噪比由12．2dB提高到了4．0dB．改善了语音质量。

Speech enhancement based on improved spectrum subtraction

Aiming at reducing musical noise and the loss of unvoiced speech in the spectral subtraction method,a reliable voice active detection algorithm which utilizes the characteristics of short-time AMDF （Average Magnitude Difference Function）and short-time AM （Average Magnitude）to detect the speech endpoint is introduced. And the spectral subtraction is improved by introducing two parameters α and β. According to collecting real speech data of noise,the outcomes of classical spectral subtraction and modified algorithm were compared. On the research of speech enhancement based on improved spectrum subtraction in the thesis,a real-time system centered on TMS320VC5509 and TLV320AIC23 Codec was built, and it increased SNR which was increased from 12.2dB to 4.0dB effectively and improved the quality of speech.