摘要
谱减法是语音增强算法中的常用算法。传统的谱减法使用直接判决法(DD)计算先验信噪比,会产生一帧的延时,不能有效滤除噪声,并且会引入"音乐噪声",影响通信效果。为了消除延时和"音乐噪声"带来的不良效果,在谱减法的基础上采用最小均方误差(MMSE)算法计算先验信噪比消除延时,并加汉明窗处理及半波整流,运用维纳滤波对带噪语音进行增强研究。结果表明,此谱减法改进算法消除了一帧的延时,并能有效滤除"音乐噪声",减小背景噪声带来的不良影响,且在主观听觉上增强了语音信号的质量,效果明显优于原始带噪语音信号。
Spectral subtraction is a commonly used algorithm in speech enhancement algorithms.Traditional spectral subtraction employs a direct decision method(DD) to compute the prior signal-to-noise ratio,and this would result in one-frame delay and couldn't filter out the noise effectively,and the introduction of "music noise" would also affect the communication effect.To eliminate the adverse effects brought about by the delay and "music noise",on the basis of spectral subtraction,the minimum mean square error(MMSE) algorithm is used to calculate the prior SNR(Signal-to-Noise Ratio),and with the addition of Hamming window and half-wave rectification,Wiener filtering is used to enhance the noisy speech.The experiment results indicate that the spectral subtraction algorithm could effectively eliminate the one-frame delay,filter out the "music noise",and reduce the adverse effects of background noise.Moreover,the quality of speech signal is enhanced on the subjective hearing,and the effect is much better than that of original speech signal.
出处
《通信技术》
2017年第9期1925-1928,共4页
Communications Technology
关键词
谱减法
MMSE
先验信噪比
半波整流
频谱修正
语谱图
spectral subtraction
MMSE
priori SNR
half-wave rectifier
spectral amendment
spectrogram