基于非线性时频掩蔽的语音盲分离方法

针对语音信号的欠定卷积混合模型,利用独立语音在时频域上的近似W-分离正交性(W-DO),提出了一种基于非线性时频掩蔽的盲分离方法。首先对多传声器观测信号在时频域上进行规范化处理,使混合信号在每个时频槽的表示与频率无关,然后采用动态聚类算法获取时频槽对应的活跃源信息,选择关于簇中心偏角的非线性函数进行时频掩蔽,从而实现语音信号的盲分离。该方法解决了经典频域盲分离算法中的频率置换问题,能有效抑制分离矩阵的空间方向扩散。仿真实验表明,与BLUES方法相比具有更优的分离性能,信噪比增益平均增加1.58 dB。     

非线性混合模式的语音盲分离算法

本文针对FIR非线性混合模型，基于最大熵算法，提出了一种以高斯混合模式概率密度函数估计替代传统对数化概率密度估计的盲分离算法，以偶函数为非线性激活函数，采用最大期望（EM）迭代算法推导了分离算法的权向量迭代公式，通过模拟仿真实验结果与传统的最大熵和高阶累积量方法比较，新算法提高了收敛速度，并有效地完成了非线性语音分离任务，抑制了干扰语音信号的影响，提高了输出信噪比。     

基于声源方位信息和非线性时频掩蔽的语音盲提取算法

针对欠定卷积混合的语音信号模型,提出一种基于声源方位信息和非线性时频掩蔽的语音盲提取算法。首先对低频段混合语音信号进行时频分析估计瞬时相对时延(ITD)并采用势函数聚类分析方法估计出声源个数及其ITD,接着锁定目标提取准确的目标语音方位信息,最后利用独立语音在时频域上的近似W一分离正交性,采用非线性时频掩蔽的方法提取目标语音。仿真实验表明,该方法能锁定任意感兴趣目标方位,能有效提取目标语音,文中实验条件下信噪比增益平均达9.5 dB。     

卷积混迭语音信号的联合块对角化盲分离方法

针对语音信号的卷积混迭模型,利用不同语音信号之间的近似独立和短时平稳特性,提出一种基于信号二阶统计量的联合块对角化方法,解决超定卷积盲分离问题。该方法采用非对角线上各子矩阵 F -范数的平方和作为联合块对角化性能的评判准则,将原四次代价函数转化为一组较为简单的二次子代价函数,每一子代价函数用于估计酉混迭矩阵的一个子矩阵。依次最小化各子函数,迭代搜索代价函数最小点,得到混迭矩阵的估计。理论分析及实验结果表明,所提方法不仅能够达到与类Jacobi经典方法同样好的分离效果,并且具有更低的计算复杂度、更快的收敛速度和对传输信道阶数、迭代初始值不敏感的特点。     

Blind convolutive separation method for speech signals via joint block diagonalization

A blind speech source separation method for the overdetermined convolutive mixture model in time-domain is proposed via joint block-diagonalization based on the mutual- independence and short-time stationarity properties of the speech signals. Taking the sum of the F-norms of all off-diagonal sub-matrices as a criterion, a novel joint block-diagonalization method is proposed to estimate the whole mixture matrix through minimizing a sequence of quadratic sub-functions corresponding to mixture sub-matrices. Both theoretical analysis and simulations show that the proposed method has much lower complexity and faster convergence speed than the classical Jacobi-like method with no performance loss. In addition, there are almost no obvious impacts of the channel order and initialization values on the convergence speed.     

时频字典学习的单通道语音增强算法

针对以往语音增强算法在非平稳噪声环境下性能急剧下降的问题,基于时频字典学习方法提出了一种新的单通道语音增强算法。首先,提出采用时频字典学习方法对噪声的频谱结构的先验信息进行建模,并将其融入到卷积非负矩阵分解的框架下;然后,在固定噪声时频字典情况下,推导了时变增益和语音时频字典的乘性迭代求解公式;最后,利用该迭代公式更新语音和噪声的时变增益系数以及语音的时频字典,通过语音时频字典和时变增益的卷积运算重构出语音的幅度谱并用二值时频掩蔽方法消除噪声干扰。实验结果表明,在多项语音质量评价指标上,本文算法都取得了更好的结果。在非平稳噪声和低信噪比环境下,相比于多带谱减法和非负稀疏编码去噪算法,本文算法更有效地消除了噪声,增强后的语音具有更好的质量。     

主观空间分离下的汉语信息掩蔽效应

基于听觉优先效应中的融合现象,探讨了主观空间分离下的汉语信息掩蔽效应。实验用左右两个扬声器来播放目标言语信号和掩蔽声音,并通过改变两个扬声器之间的延迟来操作掩蔽声音的主观空间位置。结果显示,尽管言语信号和掩蔽声音都由同样的扬声器播放而没有实际的空间分离,这种利用优先效应所产生的主观空间分离却可以提高言语识别的正确率。在信息掩蔽条件下由主观空间分离所造成的言语识别的改善显著地高于在能量掩蔽条件下的改善。这些结果为如何分离对言语信号的能量掩蔽与信息掩蔽,以及为相关的建筑声学和通讯技术的研究与应用提供了听觉心理学的参考。     

Single channel speech enhancement via time-frequency dictionary learning

A time-frequency dictionary learning approach is proposed to enhance speech contaminated by additive nonstationary noise.In this framework,a time-frequency dictionary which is learned from noise data is incorporated into the convolutive nonnegative matrix factorization framework.The update rules for the time-varying gains and speech dictionary are derived by precomputing the noise dictionary.The magnitude spectra of speech are estimated using convolution operation between the learned speech dictionary and the time-varying gains. Finally,noise is removed via binary time-frequency masking.The experimental results indicate that the proposed scheme gives better enhancement results in terms of quality measures of speech.Moreover,the proposed algorithm outperforms the multiband spectra subtraction and the non-negative sparse coding based noise reduction algorithm in nonstationary noise conditions.     

噪声背景下双输入时延混合系统的盲源分离

噪声背景下应用盲分离技术恢复源信号是盲信号处理的难点之一,本文主要研究了双输入时延有噪混合模型的盲分离方法,和传统的盲分离算法相比,该方法可以有效地利用多阵元的观测信号,对加性噪声具有相当的抑制作用。     

Blind phase retrieval and source separation of electromagnetic fields

Blind source separation of two electromagnetic fields is investigated. The difficulty of this task lies in the fact that only the power, which is the square of the sum of the electromagnetic fields, can be directly measured; the cross term of the electromagnetic fields is inevitable, and a strong correlation occurs in blind deconvolution. However, the relative phase is physically different from the field intensities, and, hence, extracting the phase during separation seems inconceivable. Our results demonstrate that the intensities and the relative phase of two electromagnetic waves can be determined with eigenvalue problem formalism even when the mixing processes are completely unknown.     

A frequency bin-wise nonlinear masking algorithm in convolutive mixtures for speech segregation

A frequency bin-wise nonlinear masking algorithm is proposed in the spectrogram domain for speech segregation in convolutive mixtures. The contributive weight from each speech source to a time-frequency unit of the mixture spectrogram is estimated by a nonlinear function based on location cues. For each sound source, a non-binary mask is formed from the estimated weights and is multiplied to the mixture spectrogram to extract the sound. Head-related transfer functions (HRTFs) are used to simulate convolutive sound mixtures perceived by listeners. Simulation results show our proposed method outperforms convolutive independent component analysis and degenerate unmixing and estimation technique methods in almost all test conditions.     

混沌信号在无线传感器网络中的盲分离

混沌信号在本质上属于非线性非高斯信号,它在无线传感器网络下的应用还涉及到信号量化问题,这使得混沌信号在此应用环境下的信号盲分离更为棘手.针对此问题,本文在容积卡尔曼粒子滤波的框架下提出一种解决方法.文中首先推导出观测信号的概率密度函数,在量化比特有限的情况下,采用最优量化器,获得最优的量化结果.在此基础上,使用容积卡尔曼滤波器产生粒子滤波中的重要性概率密度函数,融入最新的观测值,提高粒子对系统状态后验概率的逼近,提高信号盲分离的精度.仿真结果表明算法能够有效地分离混合混沌信号,参数估计的精度及其运算量均优于已有的无先导卡尔曼粒子滤波算法,其运行时间为无先导卡尔曼粒子滤波算法的88.77%.     

单观测通道船舶辐射噪声盲源分离

提出了一种适用于单观测通道的船舶辐射噪声盲源分离方法.该方法依据船舶辐射噪声远场的空间分布规律,通过将单观测通道延时和滤波的方法构造虚拟通道,使单通道转化为多通道,以实现单通道的盲源分离.仿真及实验数据分析的结果显示,分离后信号的相关系数在不同信噪比下有稳定的提高,说明该方法能在一定程度上利用单观测通道在海洋环境噪声背...     

均匀度估计的噪声源盲分离算法研究

盲信号分离技术是源输入未知时识别噪声源的一种有力手段。源的独立性常是应用盲分离算法的一个基本要求。但由于概率密度函数一般未知且估计繁琐,传统盲分离算法对源信号的独立性较难检验。为此,从信号独立性分析出发,理论上推导了随机变量的独立性和其概率分布函数的联合分布之间的关系,提出了一种独立性度量指标均匀度的估计算法,并给出了相应的盲分离算法。利用电机和海水泵的振动信号分离试验对方法进行了验证,并与现有的半熵盲分离算法进行比较,结果表明在分离效果和计算时间方面均优于现有的方法。充分说明了本文算法的有效性。     

基于噪声追踪的二值时频掩蔽到浮值掩蔽的泛化算法

虽然浮值掩蔽比二值掩蔽有更好的语音分离效果,但是由于理想浮值掩蔽难以直接估计,现有的语音分离系统通常以理想二值掩蔽估计作为计算目标。我们提出了一个二值掩蔽到浮值掩蔽的泛化算法。由于实现浮值掩蔽估计的关键在于噪声能量追踪,我们首先采用指数分布刻画以混合谱和噪声能量以混合能量及二值掩蔽为观测的条件分布。其次,采用高斯马尔柯夫条件随机场刻画噪声估计在连续几帧内的关联。最后,采用马尔柯夫链-蒙特卡洛计算噪声能量最小均方误差估计并进一步计算浮值掩蔽。实验表明,相比于基于二值掩蔽估计的常规算法,我们所提出的算法在信噪比增益和客观感知质量两方面都有显著提高。     

Real-time blind source separation system with applications to distant speech recognition

A real-time BSS system based on DUET was developed and implemented in order to assess its potential as the front-end for a DSR engine. The system uses only two closely-spaced standard omni-directional microphones and a computer soundcard and was developed for low reverberation environments with several human speakers and different noise sources.     

Blind source separation of multichannel electroencephalogram based on wavelet transform and ICA

Combination of the wavelet transform and independent component analysis （ICA） was employed for blind source separation （BSS） of multichannel electroencephalogram （EEG）. After denoising the original signals by discrete wavelet transform, high frequency components of some noises and artifacts were removed from the original signals. The denoised signals were reconstructed again for the purpose of ICA, such that the drawback that ICA cannot distinguish noises from source signals can be overcome effectively. The practical processing results showed that this method is an effective way to BSS of multichannel EEG. The method is actually a combination of wavelet transform with adaptive neural network, so it is also useful for BBS of other complex signals.     

用独立成份分析算法实现水声信号盲分离

独立成份分析算法是在研究信号盲分离过程中出现的一种新方法,本文试图将几种独立成份分析算法用于分离水声信号。分析和比较了5种算法的性能,并用仿真信号对算法进行了仿真,阐述了独立成份分析算法分离水声信号的不足。针对含噪声模型,提出了一种基于独立成份分析算法成功分离水声信号的方法。     

Comodulation masking release for speech stimuli.

This study sought to determine whether speech recognition in a modulating noise background can be facilitated by a process attributable to comodulation masking release (CMR). Experiment 1 examined the masked identification of six filtered vowels as a function of the number of comodulated noisebands present. A benefit of increased number was observed, consistent with an interpretation in terms of CMR, although it could not be certain that the basis of the discrimination was word recognition in the semantic sense. Experiment 2 made use of a forced-choice rhyming test in which the response foils differed only in a single filtered consonant; again, the measure of interest was performance as a function of the number of comodulated noisebands present. No evidence for a suprathreshold CMR was observed. Experiment 3 made use of open-set sentence material and employed a different paradigm, which allowed a measure of CMR in terms of the difference between thresholds in correlated and uncorrelated noise to be determined. While a CMR for speech detection was observed, no CMR for speech recognition was found. It was concluded that CMR is most evident in masked detection tasks and that diminishing returns are encountered as the signal-to-masker ratio is raised.