一种基于高斯分布模型的语音增强方法

对背景噪声的估计,关键是使算法能跟得上噪声变化,以及时更新噪声的估计值。为了增强对非平稳噪声的跟踪估计能力,在Imrca算法的基础上,提出了基于前向和后向最小值搜索相结合的改进算法。验证实验内容包括：用改进算法、原Imrca算法,基本谱减法对含同种噪声不同信噪比的含噪语音处理后,恢复出来的语音的对比;以及在低输入信噪比情况下,对加入babble噪声的含噪语音经改进算法、原Imrca算法、基本谱减法处理后,所得到的平均分段信噪比的对比。实验结果表明,该改进算法对含噪语音的处理效果要优于原算法,对含有非平稳的噪声的含噪语音,取得了较好的增强效果。

Speech Enhancement Algorithm Based on Imrca Using Bidirectional Search

The key to the estimation of background noise is to make the algorithm catch up with the change of noises and update its estimated value in time. In order to enchance the tracing ability for nonstationary noise, an improved algorithm based on Imrca and in combination of forward with backward searching is proposed. The experiments include the comparisons among the recovered noisy speeches with the same noise and different SNR, acquired by the improved algorithm, original Imrca algorithm and spectrum subtration, and the comarisons among the average segment SNRs in low input SNR of babble-added noisy speech acquired by these three algorithms. The experiment results show that the improved algorithm is better than others in enchancing the noisy-speech, and is even more effective in enchancing the noisy speech with nonstationary noise.