论文部分内容阅读
提出了基于噪声谱结构特性的谱减法,在不增加语音失真的情况下,抑制传统谱减法的“音乐噪声”。首先,依据噪声谱结构特性在频带间自适应平滑周期图,减小谱估计方差的同时,避免噪声非连续谱的能量泄露;其次,依据噪声谱的结构特性,对增益函数进行自适应调整以更有效的抑制有调噪声。测试结果表明,不论对宽带噪声还是对窄带噪声,本文算法在信噪比提高和噪声抑制量等客观评价指标上都明显优于传统谱减法。非正式主观测听进一步验证了本文算法的有效性。
A spectral subtraction method based on the structure of the noise spectrum is proposed, which suppresses “music noise ” of the traditional spectral subtraction without increasing the speech distortion. First of all, according to the structure of noise spectrum, adaptively smoothed period diagram in the frequency band to reduce the variance of spectral estimation and avoid the energy leakage of the discontinuous spectrum. Secondly, according to the structural characteristics of the noise spectrum, the gain function is adaptively adjusted More effective suppression of noise. The test results show that the proposed algorithm outperforms the traditional spectral subtraction method both objectively and objectively in terms of signal to noise ratio (SNR) and noise suppression, both for wideband noise and narrowband noise. Informal subjective listening further validates the effectiveness of the proposed algorithm.