多反复结构模型的精确音乐分离方法

针对基本反复模型音乐分离方法自适应性差的问题,提出一种基于美标度倒谱系数(MFCC)的多反复结构模型的音乐分离方法。首先,提取出音乐信号的MFCC系数矩阵(39维的数据构成);然后利用余弦特性得到其相似矩阵,进而将相似度一致的片段划分到一起,建立不同的反复结构模型;之后结合理想二元掩蔽(]BM)分离出背景音乐及歌声的频谱,相应的时域信号则由傅里叶逆变换获得;最后,在不同类型、长度的音乐文件上测试了算法性能,将提出的算法与Rafii的反复算法和Ozerov的灵活窗非负矩阵分解方法进行对比。实验结果表明,改进方法在分离性能上最高提高3 dB左右,并且对于曲调变换大的音乐提高效果更为明显,从而证实了改进方法是一种有效的音乐分离方法,并且更具稳定性。      

时间及通道双维序列注意力音乐声源分离方法

针对音乐中乐器声源表征特异性不足的问题,结合乐器声源与曲式内容相关的结构特征,提出了双维序列注意力/时域端到端音乐源分离方法。首先,由于各乐器声源在曲式不同部分的出现具有显著规律性,因此从时间及特征通道两个维度,对特征基函数进行差异化注意力加权。其次,在损失函数中引入频率多分辨因子,同时从时域及频域衡量分离后声源和理想声源间的差异。在MUSDB18数据集上的实验结果表明,同时强调声源的时域曲式结构特征和离散谐波特征,可以进一步改善乐器声源分离效果。与目前最先进的时域端到端音乐源分离方法 Demucs相比,信噪比指标提升了0.40 dB,且在鼓和低音等声源的分离上表现尤为突出,鼓声源信噪比指标提升0.13 dB,低音声源信噪比指标提升0.60 dB。充分利用声源的语义内容及声学特征等多维度先验知识,可以进一步提升声源表征的特异性,从而提升声源的可分离程度。     

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。      

子带t分布的快速独立向量分析在语声盲源 分离中的应用*

为了提高独立向量分析算法在盲语声分离任务中的分离性能,降低算法计算复杂度,并改善目前尚未完全解决的顺序模糊性的问题,该文提出一种基于子带t分布的快速独立向量分析算法。在声源模型方面,该算法首先利用语声信号重尾分布的特性,假设声源概率密度函数服从t分布,同时采用子带建模的方法来增强同一声源相邻频点的高阶依赖性,进而减轻频点间的顺序不一致问题。在空间模型方面,该算法采用秩1更新的方式估计声源信号,避免矩阵求逆操作和分离矩阵的估计,从而降低计算复杂度。实验结果表明,与现有的基于独立向量分析的盲源分离算法相比,该算法能够在相同的迭代次数下取得更优的语声分离性能。     

一种基于声相似律的旋转流体机械噪声声源分离方法

针对管道系统中的流体机械发声问题,提出了一种声源函数与声响应函数的频谱分离方法。该方法以声相似律为基础,分别采用最小均方值多重线性回归方法和高斯滤波获得声源函数和声响应函数。通过算例验证了该方法的有效性,辨识得到的声源函数和声响应函数的相对误差均远小于现有方法。利用该方法分析了自由空间和带管道两种安装条件下轴流风扇的辐射噪声,辨识得到了相应的声源函数、声响应函数和叶尖速度律指数,并对风扇的主要噪声成分进行了分析。      

单全息面的直接声场分离方法

为了简化分离全息面两侧相干声源的过程,提出仅依据单全息面上的数据直接进行声场分离的方法。该方法通过将等效源配置在球面上,并根据等效源法近场声全息的重构原理,建立全息面上有误差的测量值和计算值之间的数据关系,理论推导出单面声场分离方法。仿真分析干扰声源为脉动球源和受迫振动的简支薄钢板的分离结果,并对双扬声器的声场分离进行了实验验证。结果表明:该方法对两种干扰源都具有较好的分离精度,且有较高的可容忍误差。      

差分传声器阵列期望最大化多声源方位估计方法

针对小尺寸传声器阵列多声源方位估计易受混响噪声影响的问题,提出了一种适用于差分传声器阵列的期望最大化多声源方位估计方法。首先,该方法利用期望最大化算法求解出各个时频点瞬时方位估计所应满足的高斯混合模型参数;然后,通过时频点分离技术估计出各声源的方位值。针对现有的硬、软时频点分离技术应用于差分传声器阵列所存在的缺陷,还提出了一种改进的时频点分离方法,该方法融合了软、硬分离方法所具有的优点,有效降低了时频点分离结果对混响噪声的敏感性。仿真和实测实验结果表明:相较于现有的差分传声器阵列多声源估计方法,所提方法在混响噪声环境下具有更高的估计精度和稳健性能。      

利用时频相关性的球谐波阶数感知鲁棒伪声强多声源定位

针对伪声强多声源定位方法对环境噪声和混响敏感的问题,利用时频相关性构造球谐波阶数感知因子,提出了一种鲁棒的伪声强声源定位方法。与现有的球谐波阶数感知因子不同,所构造的球谐波阶数感知因子充分利用了相邻时频点中属于同一声源的特征波束间存在的相关性。理论分析表明,所构造的球谐波阶数感知因子对环境噪声和混响的抑制能力更强。仿真结果显示,与现有的阶数感知方法相比,在信噪比为10 dB、混响时间为0.4～1.0 s时,所提方法的定位精度提升了1.3°～1.9°,同时计算复杂度减小了25%。最后通过实测实验进一步验证了所提方法在实际声场环境中的有效性。     

基于总体最小二乘算法的平稳声信号二阶盲分离方法

利用声源信号相互独立的已知条件,提出分离平稳声信号的总体最小二乘二阶盲分离算法,该算法在联合近似对角化相关函数矩阵的约束下估计分离矩阵,在同时考虑估计矩阵与观测信号误差的情况下给出修正模型,在修正模型总体误差最小时分离源信号。提出的算法具有无需选择非线性函数,不存在收敛到局部最小值与计算速度快的特点,分离的声源信号具有较小的失真。半消音室电机噪声分离实验验证了上述算法的有效性。     

深海声影区时频谱干涉结构与声源定位*

被动声呐探测位于深海声影区的水面舰船辐射噪声时,接收信噪比通常较低,导致声源被动定位方法的性能较差。针对这一问题,提出一种利用接收信号时频谱干涉结构的声源距离和径向速度联合估计方法。首先根据射线声学理论,建立时频谱沿频率轴和时间轴的干涉条纹周期与声源距离和径向速度的关系,然后对接收信号时频谱进行二维傅里叶变换和多频带处理以估计上述干涉条纹周期,最后解算声源距离和径向速度。仿真和海试数据处理结果表明,相比于现有利用接收信号自相关的声源距离估计方法,该文利用时频谱二维傅里叶变换的声源定位方法具有较好的稳健性,比较适用于低信噪比条件下的声源被动定位。     

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.     

一个快速自动音乐记谱方法

自动音乐记谱是音乐信号处理中的关键技术。本文描述了一个快速的自动复音音乐记谱方法。该方法采用回声器时频分析(RTFI)作为时频分析工具,主要由两个阶段组成,能量基的音符切分和多基频估计。本文所采用的多基频估计方法首先将RTFI能量谱按照谐音组合原理转换为基频能量谱,并基于基频能量谱采用简单的峰拾起方法对基频做初步估计;然后根据频谱不规律性和乐音谐音结构的基本假定,消除初步估计中的错误预测。     

A new blind fault component separation algorithm for a single-channel mechanical signal mixture

A vibration signal collected from a complex machine consists of multiple vibration components, which are system responses excited by several sources. This paper reports a new blind component separation (BCS) method for extracting different mechanical fault features. By applying the proposed method, a single-channel mixed signal can be decomposed into two parts: the periodic and transient subsets. The periodic subset is related to the imbalance, misalignment and eccentricity of a machine. The transient subset refers to abnormal impulsive phenomena, such as those caused by localized bearing faults. The proposed method includes two individual strategies to deal with these different characteristics. The first extracts the sub-Gaussian periodic signal by minimizing the kurtosis of the equalized signals. The second detects the super-Gaussian transient signal by minimizing the smoothness index of the equalized signals. Here, the equalized signals are derived by an eigenvector algorithm that is a successful solution to the blind equalization problem. To reduce the computing time needed to select the equalizer length, a simple optimization method is introduced to minimize the kurtosis and smoothness index, respectively. Finally, simulated multiple-fault signals and a real multiple-fault signal collected from an industrial machine are used to validate the proposed method. The results show that the proposed method is able to effectively decompose the multiple-fault vibration mixture into periodic components and random non-stationary transient components. In addition, the equalizer length can be intelligently determined using the proposed method.     

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。      

Music/voice separation based on the multi-repeating structure of Mel cepstrum coefficient

For the poor adaptability of the original repeating pattern,an improved music separation method of multi-repeating structure of Mel cepstrum coefficient(MFCC) is proposed.Firstly,the MFCC coefficient matrix(39-dimensional data) of the music signal was extracted.Then the cosine characteristic was applied to the count of similarity matrix of MFCC,and the fragments with consistent similarity are putted together.Next different repeating patterns are built for different groups.Thereby the spectrums of the background music and vocal were separated combined with ideal binary masking(IBM),and the corresponding time domain signals were obtained by inverse Fourier transform.Fnally,the improved method was tested on the music database of different types and length,and the separation results were compared with repeating method of Rafii and the non-negative matrix factorization based on flexible framework method of Ozerov.The experimental results showed that the separation performance of improved method was improved about 3 dB,and the performance of music with melody changed larger was significantly improved.Experiments verified that the improved method was an effective music separation algorithm and more stability.     

舰船壳体结构噪声分离技术研究——非线性频谱特征及特征谱分离策略

为分离舰船壳体结构非线性系统的结构噪声,考虑其非线性及响应谱特征,将其理想成纯输入非线性系统,采用Volterra级数模型计算其广义频响函数,获得非线性频谱特征,提出非线性系统噪声源分析策略,并进行数值仿真和试验研究,其结果表明:非线性频谱特征合理存在;提出的非线性系统噪声源分析策略合理有效,可实现系统激励源和广义频响函数的参数估计。非线性频谱特征及提出的非线性系统噪声源分析策略,为深入开展舰船壳体结构噪声的非线性特征识别及其特征谱分离研究提供基础。      

Free radical EPR spectroscopy analysis using blind source separation

In this paper, we propose a novel approach for electron paramagnetic resonance (EPR) mixture spectra analysis based on blind source separation (BSS) technique. EPR spectrum of a free radical is often superimposed by overlapping spectra of other species. It is important and challenging to accurately identify and quantify the 'pure' spectra from such mixtures. In this study, an automated BSS method implementing independent component analysis is used to extract the components from mixed EPR spectra that contain overlapping components of different paramagnetic centers. To apply this method, there is no requirement to know the component spectra or the number of components in advance. The method is applied to analyze free radical EPR spectra which are collected from standard chemical system, cultured cell suspense, and ex vivo rat kidneys by spin trapping EPR technique. Results show that the BSS method proposed here is capable of identifying the component EPR spectra from mixtures with unknown compositions. The BSS technique can offer powerful aids in resolving spectral overlapping problems in general EPR spectroscopy analysis.     

频域盲源分离与波束形成结合抑制方向性强干扰方法

针对方向性强干扰严重影响无源声呐弱目标检测的问题,提出了频域盲源分离与波束形成结合的干扰抑制方法:以子带分解的方法实现宽带干扰抑制。对每个子带进行频域盲源分离,并估计出各分离信号的方位,将与给定强干扰方位匹配的分离信号置零,利用估计的解混矩阵和处理后的分离信号重构回阵元域信号并进行波束形成实现目标方位估计。声呐模拟器数据与海试数据验证结果表明,相对于传统零陷常规波束形成与零陷最小方差无失真响应波束形成方法有2 dB以上的增益,约6 dB的背景级降低,证明该方法在抑制方向性强干扰方面是有效的。