论文部分内容阅读
为降低噪声对语音通信的干扰,提出了一种基于谱减的语音增强算法的改进方法。根据噪声频谱的G auss统计模型修正语音增强过程中噪声频谱的估计方法,利用帧内、帧间约束估计每一个频点的先验信噪比,提出了一种简便的估计语音在每一个频点出现的概率的方法,得出了修正的语音增强算法。在白噪声以及坦克噪声环境中分别对算法的性能进行测试。实验结果表明:在几乎不损伤语音清晰度的前提下,该算法使“音乐噪声”得到了更好的抑制,同时信噪比提高了约8 dB以上。
In order to reduce the interference of noise on voice communication, an improved method based on spectral subtraction is proposed. According to the G auss statistical model of noise spectrum, the method of estimating the noise spectrum in speech enhancement process is modified, and the a priori SNR of each frequency point is estimated by intra and inter frame constraints. A simple estimate of the speech at each frequency Point probability of emergence of the method, a revised speech enhancement algorithm. The performance of the algorithm was tested in white noise and tank noise environment respectively. The experimental results show that the proposed algorithm can restrain “music noise” better and improve signal-to-noise ratio more than 8 dB at the same time without impairing speech intelligibility.