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。      

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.     

联合深度编解码网络和时频掩蔽估计的单通道语音增强

提出了一种联合深度编解码神经网络和时频掩蔽估计的语音增强方法。该方法利用深度编解码网络估计时频掩蔽表示,并联合带噪语音的幅度谱学习带噪语音与纯净语音幅度谱之间的非线性映射关系。深度编解码网络采用卷积-反卷积网络结构。在编码端,利用卷积网络的局部感知特性,对带噪语音的时频域结构特征进行建模,提取语音特征,同时抑制背景噪声。在解码端,利用编码端提取到的语音特征逐层恢复局部细节信息并重构语音信号。同时,在编解码端对应层之间引入跳跃连接,以减少由于池化和全连接操作导致的低层细节信息丢失的问题。在TIMIT语音库和不完全匹配噪声集下进行仿真实验,实验结果表明,该方法可以有效抑制噪声,且能较好地恢复出语音细节成分。      

Time-frequency analysis of time-varying modulated signals based on improved energy separation by iterative generalized demodulation

Energy separation algorithm is good at tracking instantaneous changes in frequency and amplitude of modulated signals, but it is subject to the constraints of mono-component and narrow band. In most cases, time-varying modulated vibration signals of machinery consist of multiple components, and have so complicated instantaneous frequency trajectories on time-frequency plane that they overlap in frequency domain. For such signals, conventional filters fail to obtain mono-components of narrow band, and their rectangular decomposition of time-frequency plane may split instantaneous frequency trajectories thus resulting in information loss. Regarding the advantage of generalized demodulation method in decomposing multi-component signals into mono-components, an iterative generalized demodulation method is used as a preprocessing tool to separate signals into mono-components, so as to satisfy the requirements by energy separation algorithm. By this improvement, energy separation algorithm can be generalized to a broad range of signals, as long as the instantaneous frequency trajectories of signal components do not intersect on time-frequency plane. Due to the good adaptability of energy separation algorithm to instantaneous changes in signals and the mono-component decomposition nature of generalized demodulation, the derived time-frequency energy distribution has fine resolution and is free from cross term interferences. The good performance of the proposed time-frequency analysis is illustrated by analyses of a simulated signal and the on-site recorded nonstationary vibration signal of a hydroturbine rotor during a shut-down transient process, showing that it has potential to analyze time-varying modulated signals of multi-components.     

基于时频分析的多目标盲波束形成算法

针对现有盲波束形成算法适用范围较窄, 多目标信号分离级联模式结构复杂、并联模式稳定性较差等问题, 提出一种基于时频分析的多目标盲波束形成算法. 该算法首先利用时频分析技术给出信号导向矢量的不确定集, 然后优化求解导向矢量的最优估计, 最后利用Capon方法实现多目标信号的并行输出. 理论分析及仿真结果表明, 该算法对信号特性没有特殊要求, 适用性较广, 性能稳定, 且输出信干噪比高于其他盲波束形成算法, 接近于最优Capon波束形成器.     

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.     

水下目标回波盲分离性能的瞬时频率特征评价方法

在水下主动声呐目标回波与混响盲分离中,针对分离结果顺序不确定性以及缺乏分离有效性衡量手段的问题,提出了以信号瞬时频率特征为指标的盲分离性能评价方法。推导了目标回波与混响的时频分布特性,理论表明目标回波在瞬时频率序列的中心偏离程度以及整体随机程度上低于混响,据此提取信号的瞬时频率方差与瞬时频率熵两种信号特征,并将二者作为从盲分离结果中识别目标回波的依据。海试数据结果表明,在盲分离得到的所有分离信号中,目标回波具有最小的瞬时频率特征值,并且该特征值越小,目标回波与混响的盲分离程度就越高。      

基于多项式调频Fourier变换的信号分量提取方法

为了从含有噪声的混合信号中有效提取各个信号分量, 提出一种基于多项式调频Fourier变换的分量提取方法. 通过研究Fourier变换和分数阶Fourier变换的信号能量积累方式及变换基函数的时频表示, 提出利用时频平面上的多项式调频曲线族代替Fourier变换和分数阶Fourier 变换的调频直线族, 将变换的适用范围扩展到非线性调频信号. 采用粒子群智能优化算法搜索调频曲线族的最优多项式参数, 使混合信号中的某一分量在多项式调频Fourier域上能量谱集中. 最后对能量谱集中的分量进行窄带滤波, 并利用多项式调频逆Fourier变换重构信号分量. 仿真实验结果表明, 该方法不仅能够提取混合信号中的线性调频分量, 还能够实现非线性调频分量的能量谱集中、信号分离和时频特征提取.     

Blind single-channel suppression of late reverberation based on perceptual reverberation modeling

A blind method for suppressing late reverberation from speech and audio signals is presented. The proposed technique operates both on the spectral and on the sub-band domains employing a single input channel. At first, a preliminary rough clean signal estimation is required and for this, any standard technique may be applied; however here the estimate is obtained through spectral subtraction. Then, an auditory masking model is employed in sub-bands to extract the reverberation masking index (RMI) which identifies signal regions with perceived alterations due to late reverberation. Utilizing a selective signal processing technique only these regions are suppressed through sub-band temporal envelope filtering based on analytical expressions. Objective and subjective measures indicate that the proposed method achieves significant late reverberation suppression for both speech and music signals over a wide range of reverberation time (RT) scenarios.     

噪声情况下采用稀疏非负矩阵分解与深度吸引子网络的人声分离算法

为实现噪声情况下的人声分离,提出了一种采用稀疏非负矩阵分解与深度吸引子网络的单通道人声分离算法。首先,通过训练得到人声与噪声的字典矩阵,将其作为先验信息从带噪混合语音中分离出人声与噪声的系数矩阵;然后,根据人声系数矩阵中不同的声源成分在嵌入空间中的相似性不同,使用深度吸引子网络将其分离为各声源语音的系数矩阵;最后,使用分离得到的各语音系数矩阵与人声的字典矩阵重构干净的分离语音。在不同噪声情况下的实验结果表明,本文算法能够在抑制背景噪声的同时提高分离语音的整体质量,优于结合声噪人声分离模型的对比算法。      

一种基于模式识别的多路盲语音提取方法

盲分离算法能在缺少混合系统参数的条件下仅由观测信号估计初始源,但分离信号存在固有的排列模糊性,这往往导致两次批处理过程中同一信号"对不准",因此很难获得连续的源信号。本文针对盲声源分离中存在的相同问题,根据语音和其他音频信号的特征差异,提出一种修正的自相关函数并以其值作为一个特征基元来表征声音信号的时序相关特性,同时用平均声门波形状参数作为另一个特征基元来表征语音产生的生理效应。以这两个参数作为识别不同音频信号的二维模式特征,采用一种模糊聚类算法提取多路盲分离语音。本方法有效克服了批处理盲声源分离中的信号排列顺序的不确定性,并通过选择合适的阈值提取多路连续语音。仿真给出了5路混合音频信号中盲提取两路连续语音的实验结果。     

联合精确比值掩蔽与深度神经网络的单通道语音增强方法

针对目前有监督语音增强忽略了纯净语音、噪声与带噪语音之间的幅度谱相似性对增强效果影响等问题,提出了一种联合精确比值掩蔽(ARM)与深度神经网络(DNN)的语音增强方法。该方法利用纯净语音与带噪语音、噪声与带噪语音的幅度谱归一化互相关系数,设计了一种基于时频域理想比值掩蔽的精确比值掩蔽作为目标掩蔽;然后以纯净语音和噪声幅度谱为训练目标的DNN为基线,通过该DNN的输出来估计目标掩蔽,并对基线DNN和目标掩蔽进行联合优化,增强语音由目标掩蔽从带噪语音中估计得到;此外,考虑到纯净语音与噪声的区分性信息,采用一种区分性训练函数代替均方误差(MSE)函数作为基线DNN的目标函数,以使网络输出更加准确。实验表明,区分性训练函数提升了基线DNN以及整个联合优化网络的增强效果;在匹配噪声和不匹配噪声下,相比于其它常见DNN方法,本文方法取得了更高的平均客观语音质量评估(PESQ)和短时客观可懂度(STOI),增强后的语音保留了更多语音成分,同时对噪声的抑制效果更加明显。      

基于谐波和能量特征的单声道浊语音分离方法

提出了一种基于谐波和能量特征的单声道浊语音分离方法。该方法将浊语音分离问题转化为声音在时频域的分类问题。首先,在已有谐波特征的基础上,引入能量特征。然后,对于谐波特征明显且能量大的时频单元,在分类器训练阶段复制它们的特征。实验结果表明该方法相比之前的方法有更好的信噪比增益。通过引入能量特征和特征复制,改善了浊语音的分离效果。      

Solving permutations in frequency-domain for blind separation of an arbitrary number of speech sources

Blind separation of speech sources in reverberant environments is usually performed in the time-frequency domain, which gives rise to the permutation problem: the different ordering of estimated sources for different frequency components. A two-stage method to solve permutations with an arbitrary number of sources is proposed. The suggested procedure is based on the spectral consistency of the sources. At the first stage frequency bins are compared with each other, while at the second stage the neighboring frequencies are emphasized. Experiments for perfect separation situations and for live recordings show that the proposed method improves the results of existing approaches.     

Isolating the energetic component of speech-on-speech masking with ideal time-frequency segregation

When a target speech signal is obscured by an interfering speech wave form, comprehension of the target message depends both on the successful detection of the energy from the target speech wave form and on the successful extraction and recognition of the spectro-temporal energy pattern of the target out of a background of acoustically similar masker sounds. This study attempted to isolate the effects that energetic masking, defined as the loss of detectable target information due to the spectral overlap of the target and masking signals, has on multitalker speech perception. This was achieved through the use of ideal time-frequency binary masks that retained those spectro-temporal regions of the acoustic mixture that were dominated by the target speech but eliminated those regions that were dominated by the interfering speech. The results suggest that energetic masking plays a relatively small role in the overall masking that occurs when speech is masked by interfering speech but a much more significant role when speech is masked by interfering noise.     

时频域独立元分析方法实现超声兰姆波的多模式分离

为方便兰姆波信号分析与模式定征,提出一种将短时傅里叶变换(Short-Time Fourier Transform,STFT)与独立元分析(Independent Component Analysis,ICA)相结合的多模式超声兰姆波识别方法.首先通过STFT将兰姆波时域信号投影至时频域,基于各模式信号在时频域相对独立...     

一种基于人工蜂群算法的混沌信号盲分离方法

混沌信号所固有的非周期、宽带频谱和对初值极度敏感等特性使得对这类信号进行盲分离极为困难. 针对这一问题, 提出一种新的盲分离方法, 该方法通过相空间重构来构造代价函数, 将混沌信号的盲分离转化为一个无约束优化问题, 并利用人工蜂群算法进行求解. 不同于现有的独立成分分析方法仅使用混合信号的统计特性来解决分离问题, 该方法能充分利用混合信号内在的动态特性, 因而在处理混沌信号这种确定性信号时能获得更好的分离效果. 此外, 正交矩阵的参数化表示有效地降低了盲分离问题的复杂性, 使优化过程能快速收敛. 实验结果表明, 该方法具有较快的收敛速度和较高的数值精度, 在分离混沌信号时其整体性能优于现有的几种盲分离方法. 同时, 在分离混沌-高斯混合信号的实验中该方法也展现出优异良好的性能, 这表明该文的方法有应用潜力.     

结合区分性训练深度神经网络的歌声与伴奏分离方法

针对音乐信号中的歌声与伴奏相互关联难以分离的问题,提出了一种区分性训练深度神经网络(Deep Neural Network,DNN)的音乐分离方法。首先,在DNN模型的基础上同时考虑歌声与伴奏间的重建误差和区分性信息,提出了一种改进的目标函数进行区分性训练;其次,在DNN模型上额外添加一层,引入时频掩蔽对估计出的歌声伴奏进行联合优化,相应的时域信号由傅里叶逆变换获得;最后,验证不同参数设置对分离性能的影响,并与现有的音乐分离方法进行对比.实验结果表明,改进的目标函数和时频掩蔽的引入明显提高了DNN的分离性能,且与现有的音乐分离方法相比分离性能最高提高了4 dB从而证实所提方法是一种有效的音乐分离方法。      

A separation method of singing and accompaniment combining discriminative training deep neural network

For the difficulty of separation between singing and accompaniment in the musical signals,an improved music separation method of based on discriminative training depth neural network(DNN) was proposed.Firstly,based on the DNN model,considering the reconstruction errors and discrimination information between singing and accompaniment,an improved objective function was presented to discriminate the training;Then,an additional layer was added to DNN model,introducing the time-frequency masking to optimize the estimated accompaniment of the song,and the corresponding time-domain signal was obtained by inverse Fourier transform;Finally,the influence of different parameters on the separation performance was verified,and compared it with the existing music separation methods.The experimental results showed that the improved objective function and the introduction of time-frequency masking significantly improved the separation performance of the DNN,and the separation performance was improved about 4 dB compared with other existing music separation methods,thus verifying that the proposed method was an effective music separation algorithm.     

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。