语音信号中,噪声的存在极大地影响了语音质量,必须采用语音增强技术在一定程度上抑制背景噪声,最大限度地改善语音通信质量,提高语音的可懂度。语音增强就是从带噪语音的信号中提取尽可能纯净的原始语音信号。谱减法由于其计算简单,实时性好,所以得到了广泛应用。文中介绍了基于谱减法的语音增强算法的基本原理,并针对传统谱减法残余"音乐噪声"过强的缺点,分析了两种典型的改进谱减法的原理。实验结果表明改进算法更好地抑制了音乐噪声,有效提高了语音增强效果,改善了语音质量。

For the speech signal,voice quality is greatly affected by the noise. Speech enhancement techniques must be used to suppress background noise,improving voice quality and voice clear degrees. The speech enhancement is to get the speech as original as possible from the speech signal with noise. The spectral subtraction speech enhancement is utilized broadly because it is simple and easy for the re-al-time processing. In this paper,the basic principle of algorithm of the spectral subtraction is presented. Aiming at reducing annoying musical noise in the standard spectral subtraction algorithm,the principle of two kinds of typical improved spectral subtraction is ana-lyzed. Experimental results show that the proposed algorithm reduces the musical noise effectively and improves the speech quality.

为了使得MELP声码器在高噪声环境下仍然获得较好的语音效果,需对含噪声语音进行语音增强。本文采用谱减法和独立分量分析相结合方法,对语音进行增强。该方法可以在不增加语音采样硬件的条件下,满足独立分量分析中观测信号的数目不少于源信号数目的约束条件。结果表明,该方法能较好的分离出噪声和语音信号,增强输入到MELP声码器中的语音信号,提高MELP声码器在高噪声环境下应用的语音效果。

To get better hearing performance in strong noise environment,MELP vocoder should have the ability to reduce the noise mixed into the speech. The method combining spectral subtraction technique with independent component analysis is employed for noise reduction in this paper. The restructured speech and the original noised speech act as the observed signals to fulfil the constraint condition of ICA that is the number of observed signals must be greater than or equal to source signals.Then,the ICA algorithm is applied to get purer speech from original noised speech. The presented method can obey the constraints for application of ICA,without increasing the complexity of data acquisition circuit.Simulation results show that good effect of noise reduction is achieved,and the presented method can be applied to MELP vocoder used in strong noise environment to improve the speech quality without additional hardware.

针对固定阈值小波包语音增强算法在去噪时会损失语音信号的问题，文中提出了一种新的自适应阈值小波包语音增强算法。该算法先利用带噪语音的小波包变换系数估计出后验信噪比，再由含有后验信噪比因子的sigmoid函数作为平滑因子对随尺度变化的阈值进行相邻帧的平滑，最后由后验信噪比自适应修正平滑阈值，减少语音失真；仿真实验结果表明，该算法在去噪的同时减少了语音信号的损失，有效地提高了增强语音的信噪比和分段信噪比，较固定阈值小波包语音增强算法具有明显的优越性。

Aiming at the problem of the fixed threshold wavelet packet speech enhancement algorithm loses the speech signal in denoising, this paper proposed a new adaptive threshold of wavelet packet for speech enhancement algorithm. Firstly, the algo-rithm uses the wavelet packet coefficients of noisy speech to estimate the posteriori SNR. Secondly, the sigmoid function con-tains a posteriori the signal-to-noise ratio (SNR) factor as a smooth factor smoothes adjacent frame threshold with scale chang-ing. Finally, the posterior SNR adaptively adjusts the smoothed threshold, which reduced speech distortion. Simulation experi-mental results show that this algorithm in denoising reduces the loss of the speech signal, effectively improves the SNR and SS-NR of the enhanced speech, and has obvious superiority than a fixed threshold wavelet packet speech enhancement algorithm.

提出一种具有较高可懂度的基于维纳滤波的语音增强算法。相比于其他语音增强算法,维纳滤波法可以明显提高语音质量且含有较少的音乐噪声,但是它和其他现有语音增强算法一样,都无法有效提高语音可懂度。因为维纳滤波法和其他现有算法都过多注重噪声减少,却忽略了SNR（信噪比）的估计误差和不同的语音幅度谱畸变对可懂度有更重要的影响。为改进这些缺点,此研究依据于先验SNR和增益函数来判定SNR估计误差和语音畸变区域,然后对先验SNR小于-10 d B区域的增益函数进行修正,以及幅度谱畸变大于6.02 d B区域语音进行限制。实验证明,该算法能有效提升增强后语音可懂度NCM（归一化协方差方法）的评测值。

We propose a Wiener filtering-based speech enhancement algorithm with higher intelligibility.Compared with other speech en-hancement algorithms, the Wiener filtering can significantly improve the quality of speech but contains less musical noise, however the same as them, it is unable to effectively improve speech intelligibility either.It is because that the Wiener filtering and other existing algorithms pay too more attention to noise reduction while ignoring that the SNR ( signal-to-noise ratio) estimation errors and the different speech magnitude spectrum distortions have a greater influence on speech intelligibility.In order to make up these shortages, in our study we determine the SNR estimation errors and the speech distortions area relying on priori SNR and gain function, and then modify the gain functions in the areas where the priori SNR is less than -10 dB, and constrain the speech in the areas where the distortion of amplification spectrum is beyond 6.02 dB.Experi

提出一种可适应非平稳噪声环境的基于码本学习的改进谱减语音增强算法。该算法分为训练阶段和增强阶段。训练阶段，使用自回归模型对语音和噪声的频谱形状进行建模并构造语音和噪声码本；增强阶段，采用对数谱最小化算法估计出语音和噪声的频谱，通过谱相减消除噪声。算法在每个时间帧估计语音和噪声频谱，即使在语音存在时仍能够有效跟踪快速变化的非平稳噪声；采用自回归模型能得到噪声频谱的平滑估计，减少了音乐噪声。实验仿真表明，相比于传统谱减法和多带谱减法，改进的谱减法具有更好的噪声抑制性能并且语音失真更小。

An improved spectral subtraction algorithm based on codebook learning for speech enhancement in non-stationary noise conditions is proposed. The proposed algorithm contains two steps. The priori information about the spectrum of speech and noise is modeled using autoregressive model and the speech and noise codebooks are constructed. The speech and noise are estimated in each time frame by solving a log-spectral distortion minimization problem. The proposed algorithm can adapt to varying levels of noise even while speech is present. On the other hand, autoregressive modeling results in smooth frequency spectrums and thus reduces musical noise. Experimental results show that the proposed algorithm outperforms the traditional spectral subtraction algorithm and multiband spectral subtraction algorithm.