【作者】 安扣成；

【导师】 张玲华；

【作者基本信息】 南京邮电大学 ， 信号与信息处理， 2012， 硕士

【摘要】 语音增强的主要目的是消除带噪语音信号中的噪声,提取较为纯净的语音信号,该技术对提高数字助听器的性能具有重要意义。本文主要研究数字助听器中的语音增强技术,通过对基于先验信噪比估计算法的语音增强方法和传统的基于听觉掩蔽效应的语音增强方法的研究,提出了改进算法。论文的主要工作包括:1、对数字助听器中的语音增强技术进行了研究,重点研究了基于先验信噪比估计算法的语音增强方法和传统的基于听觉掩蔽效应的语音增强方法,对不同算法进行了实验仿真和性能分析。研究发现,采用先验信噪比估计算法计算的相邻帧之间增益函数的取值变化过快,导致增强后的语音频谱存在随机尖峰,造成音乐噪声;采用传统谱减法对带噪语音信号进行初步增强造成掩蔽阈值误差偏大,对谱减系数的计算产生较大误差,使语音增强效果降低。2、为了减小音乐噪声的影响,提出了基于自适应先验信噪比估计和增益函数平滑相结合的方法,利用带噪语音信号的频谱和估计噪声频谱的差异度对增益函数进行平滑。实验表明,本文算法可以使相邻帧之间增益函数的取值缓慢变化,能够有效消除增强语音频谱上的随机尖峰,使音乐噪声得到有效抑制。3、通过对人耳听觉掩蔽效应的研究,对传统的基于听觉掩蔽效应的语音增强技术进行了改进,该方法在初步增强中采用了基于谱熵的改进谱减法,然后结合人耳听觉掩蔽效应,利用初步增强语音计算出掩蔽阈值,得到较为准确的掩蔽阈值,用以动态调整谱减参数。实验表明,该方法能有效改善语音增强的效果。更多还原

【Abstract】 The main purpose of speech enhancement is to eliminate the noise in noisy speech, to extract more pure speech and it has great significance to improve the performance of digital hearing aids. The major study of this thesis is the speech enhancement technology of digital hearing aids. This thesis studies the speech enhancement based on a prior SNR estimation and the speech enhancement based on human auditory masking effect, and proposes the improved algorithms. The main work of the thesis includes the following aspects:First, this thesis studies the speech enhancement of digital hearing aids, and does research on the speech enhancement based on a prior SNR estimation and speech enhancement based on human auditory masking threshold. And then this thesis makes simulation and performance analysis on different algorithm. Therefore we find that the gain function value between adjacent frames changes rapidly, which leads to random spikes in enhanced speech spectrum and results in musical noise. The traditional spectral subtraction has a bad effect on initial speech enhancement, resulting in a large error in masking threshold which causes to calculate the spectral subtraction coefficient inaccuracy; so it reduces the perceptual of the speech enhancement.Second, in order to reduce the impact of musical noise, an improved algorithm that combines the adaptive a prior SNR estimation with gain function smoothing is proposed. This proposed algorithm uses the difference degree of noisy speech spectrum and noise spectrum estimated to smooth gain function. The simulation results show that the proposed method can make the slowly varying values of the gain function between adjacent frames; so that it can eliminate the random spikes, suppress the musical noise effectively.Third, this thesis conducts research on the human auditory masking effect, analyzes the method based on human auditory masking effect and proposes an improved algorithm. The improved spectral subtraction based on entropy is applied. And this thesis uses the initial enhancement speech to calculate the masking threshold and obtains the more accurate threshold. So we can use the threshold to adjust the spectral subtraction coefficient. The simulation results show that the proposed method can improve the perceptual of the speech enhancement effectively.更多还原