| 摘要: |
| 本文研究了基于先验信噪比估计的维纳滤波增强算法。针对基于先验信噪比估计的维纳滤波法未能有效提高语音可懂度的缺点进行了研究与改进：首先，引入了以MMSE为准则的两步先验信噪比估计取代原“直接判决”法；其次，针对信噪比为负的区域，增益函数的过高估计会严重降低语音可懂度的问题对增益函数进行了改进，放大了负信噪比区域的噪声谱，从而降低增益函数被过高估计的可能性；最后，针对两步先验信噪比估计引起的畸变问题，在两步先验信噪比估计的基础上提出了谐波重构降噪方法，使用畸变的信号重构丢失的谐波成分，从而降低了信号的畸变。实验结果表明，改进的方法能够有效的提高语音可懂度并降低语音信号的畸变。 |
| 关键词: 数字助听器 维纳滤波 先验信噪比 谐波重构 语音可懂度 |
| DOI： |
|
| 基金项目: |
|
| Speech Enhancement based on Harmonic reconstruction filter Used in Digital Hearing Aids |
|
| () |
| Abstract: |
| This paper focuses on the speech enhancement based on a priori SNR estimation Wiener Filtering. The methodbased on priori SNR estimation Wiener Filteringcannot effectively improve the speech intelligibility. Some research is done to solve this disadvantage. Firstly,this paper uses the MMSE-based two-step a priori SNR estimate approach to replace the “decision directed” method. Secondly, the gain function ismodifiedbecause of the problem that the gain function overestimation errors in frequency bins with negative SNR are particularly harmful to speech intelligibility. Finally , to solve the distortion problem caused by two step prior SNR estimation , harmonic regeneration noise reduction method was proposed , in which distortion signal was used to reconstruct lost harmonic component . The comparative experiments with the original method demonstrate the effectiveness of the improved method and the reduction of distortion of speech signal. |
| Key words: digital hearing aids wiener filtering priori SNR harmonic regeneration speech intelligibility |