基于差分变换的语音信号压缩感知

信号在某种变换下可以稀疏表示是压缩感知研究的先验条件,正交傅里叶变换则是应用非常广泛的一种稀疏变换.但是,由于语音信号是准周期信号,对其进行傅里叶变换会造成频谱泄漏,因而引起信号重构性能的降低.本文基于语音信号准周期性的特点,提出了一种基于差分变换的语音稀疏化变换矩阵,在此基础上采用OMP优化算法来重构语音信号.实验表明,与采用正交傅里叶变换方法对语音信号进行稀疏化变换、OMP算法对语音信号进行重构的方法相比,差分变换方法的性能明显优于正交傅里叶变换的方法,即在相同重构性能时,差分变换的压缩比小于正交傅里叶变换,因而差分变换的方法大大提高了信号的压缩性能.PESQ对重构语音质量评测的结果表明差分变换方法重构的语音信号MOS得分较高,这也说明对于语音信号这一特殊信号,差分变换法具有很大的优越性.     

语音重构的DCT域加速Landweber迭代硬阈值算法

重构信号的最基本理论依据是该信号在某个变换域是稀疏的或近似稀疏的。基于语音信号在DCT域的近似稀疏性,可以采用压缩感知(Compressed Sensing,CS)理论对其进行重构。压缩感知理论中的迭代硬阈值(Iterativehard thresholding,IHT)算法以其较好的性能被广泛用来重构信号,但其收敛速度比较慢,如何提高收敛速度,一直是迭代硬阈值算法研究的重点之一。针对压缩感知理论中的IHT算法收敛速度相当慢的问题,提出了语音重构的DCT域加速Landweber迭代硬阈值(Accelerated Landweber iterative hard thresholding,ALIHT)算法。该算法对原始语音信号做DCT变换,然后在DCT域将每一步Landweber迭代分解为矩阵计算和求解两步,通过修改其中的矩阵计算部分实现Landweber迭代加速,最后通过迭代硬阈值对信号做阈值处理。实验结果表明,加速Landweber迭代硬阈值算法加快了收敛速度、减少了计算量。     

基于匹配追踪算法的电能质量研究

为了解决电能质量信号采集数据量大的问题,提出基于匹配追踪重构算法的压缩感知方法,并首次应用于电能质量信号压缩采样研究。文中通过采用不同的稀疏基和重构算法的方法,来提高原始电能质量信号重构效果。当采样数据空间稀疏基分别选取傅里叶变换基和小波变换基,重构算法分别采用正交匹配追踪(OMP)和压缩采样匹配追踪(CoSaMP)时,仿真结果表明,压缩采样比为20%时,两种重构算法的均方误差都低于3%,重构信噪比大于30dB,为电能质量信号压缩采样研究提供了一种新的思路。     

基于正交匹配追踪算法的语音信号重构研究

压缩感知理论是近年来提出的一种新兴的基于信号稀疏性的采样理论。正交匹配追踪算法是其中一种典型的重构方法,文中针对语音信号重构中存在的不足,采用正交匹配追踪算法对语音信号进行信号重构,相比于传统的压缩感知的重构算法更加地适用于对含噪语音、重构语音质量会更高,去噪效果也会更明显。为语音信号CS性能的基础性的研究提供了参考。     

行阶梯观测矩阵、对偶仿射尺度内点重构算法下的语音压缩感知

基于语音信号在离散余弦域上的近似稀疏性,针对采用随机高斯观测矩阵及线性规划方法进行语音压缩感知与重构时,重构零(近似零)系数定位能力差而导致重构效果不好的缺点,本文提出一种新的行阶梯矩阵做观测矩阵,用对偶仿射尺度内点重构算法对语音进行压缩感知与重构,并对该算法下的重构性能进行理论分析.语音压缩感知仿真结果表明,在离散余弦基下,压缩比(观测序列与原始序列样值数之比)为1∶4时,行阶梯观测矩阵下的平均重构信噪比比随机高斯观测矩阵下提高9.73dB,平均MOS分比随机高斯观测矩阵下提高1.22分.     

基于近似KLT域的语音信号压缩感知

压缩感知是近年来兴起的研究热点,该文基于语音信号在KLT域的稀疏特性,提出了基于模板匹配的近似KLT,并在基于模板匹配近似KLT域上研究了语音信号的压缩感知性能。首先验证语音信号在基于模板匹配近似KLT域上的稀疏性,然后由语音信号与观测矩阵构造相应的观测,采取固定分配每帧观测个数和按帧能量自适应分配每帧观测个数两种方案,再以观测为已知条件利用L1优化算法重构语音信号在基于模板匹配近似KLT域的稀疏系数向量,进而重构原始语音信号。实验表明,语音信号在基于模板匹配的近似KLT域的压缩感知性能较好。     

认知WSN中基于能量有效性自适应观测的梯度投影稀疏重构方法

针对认知无线传感器网络中传感器节点侧的模拟信息转换器对本地感知数据进行稀疏表示与压缩测量,该文提出一种基于能量有效性观测的梯度投影稀疏重构(GPSR)方法。该方法根据事件区域内认知节点对实际感知到的非平稳信号空时相关性结构,映射到小波正交基级联字典进行稀疏变换,通过加权能量子集函数进行自适应观测,以能量有效的方式获取合适的观测值,同时对所选观测向量进行正交化构造测量矩阵。汇聚节点采用GPSR算法进行自适应压缩重构。仿真比较了GPSR自适应重构与正交匹配追踪(OMP)重构算法。仿真结果表明,在压缩比小于0.2的区域内,基于能量有效性观测的GPSR自适应重构效果优于传统随机高斯测量信号重构。在相同节点数情况下,GPSR自适应压缩重构方法在低信噪比区域内具有较小的重构均方误差,且该方法所需观测数明显低于随机高斯观测,同时有效保障了感知节点的能耗均衡。     

帧间自适应语音信号压缩感知

近年来提出的压缩感知是一种以低于传统奈奎斯特速率对信号采样可得到精确恢复的理论。该理论很快应用于简化传统的采样硬件、缩短采样时间、以及减少数据的存储空间。针对语音信号的传输问题,本文提出一种帧间自适应语音信号压缩感知的方法。在离散余弦变换域的语音信号具有稀疏性的前提下,以大量语音信号帧的分析统计为依据,提出一种基于语音帧能量分级和帧间位置惯性的语音信号自适应压缩感知算法。实验结果表明,能量自适应可以显著地提高语音信号的恢复质量,而位置自适应可以明显地减少语音信号的恢复时间,从而本文提出的算法可以用较少的恢复时间获得较好的恢复效果。     

基于分块自相关的子空间追踪语音压缩感知

在运用压缩感知基本原理对信号重构时,针对其中正交匹配追踪（OMP）算法和子空间追踪（SP）算法的各自特点,并与自相关思想结合,提出了一种改进的分块自相关的子空间追踪（BASP）算法。实验结果表明：在相同的压缩比下,BASP语音重构算法与SP语音重构算法相比具有较好的信噪比以及MOS评分。     

基于压缩感知观测序列倒谱距离的语音端点检测算法

本文基于语音信号在离散余弦基上的近似稀疏性,采用稀疏随机观测矩阵和线性规划重构算法对语音信号进行压缩感知与重构.研究了语音信号的压缩感知观测序列特性,根据语音帧和非语音帧压缩感知观测序列频谱幅度分布分散且差异较大的特性,提出基于压缩感知观测序列倒谱距离的语音端点检测算法,并对4dB-20dB下的带噪语音进行端点检测仿真实验.仿真结果显示,基于压缩感知观测序列倒谱距离的语音端点检测算法与奈奎斯特采样下语音的倒谱距离端点检测算法一样具有良好的抗噪性能,但由于采用压缩采样,减少了端点检测算法的运算数据量.     

A COMPRESSION ALGORITHM FOR ECG BASED ON INTEGER LIFTING SCHEME WAVELET TRANSFORM

In view of the shortcomes of conventional ElectroCardioGram (ECG) compression algo- rithms,such as high complexity of operation and distortion of reconstructed signal,a new ECG compression encoding algorithm based on Set Partitioning In Hierarchical Trees (SPIHT) is brought out after studying the integer lifting scheme wavelet transform in detail.The proposed algorithm modifies zero-tree structure of SPIHT,establishes single dimensional wavelet coefficient tree of ECG signals and enhances the efficiency of SPIHT-encoding by distributing bits rationally,improving zero-tree set and ameliorating classifying method.For this improved algorithm,floating-point com- putation and storage are left out of consideration and it is easy to be implemented by hardware and software.Experimental results prove that the new algorithm has admirable features of low complexity, high speed and good performance in signal reconstruction.High compression ratio is obtained with high signal fidelity as well.     

利用离散余弦变换的语音信号压缩方案

利用离散余弦变换后的语音信号能量主要集中在低频段的特点以及语音信号的短时平稳性，研究三种基于一维离散余弦变换的语音压缩方案。Matlab仿真结果表明三种方案的数据压缩率高，重建语音信号具有艮好的清晰度和自然度。     

Discrete cosine harmonic wavelet transform and its application to signal compression and subband spectral estimation using modified group delay

This paper proposes a new harmonic wavelet transform (HWT) based on discrete cosine transform (DCTHWT) and its application for signal or image compression and subband spectral estimation using modified group delay (MGD). Further, the existing DFTHWT has also been explored for image compression. The DCTHWT provides better quality decomposed decimated signals, which enable improved compression and MGD processing. For signal/image compression, compared to the HWT based on DFT (DFTHWT), the DCTHWT reduces the reconstruction error. Compared to DFTHWT for the speech signal considered for a compression factor of 0.62, the DCTWHT provides a 30% reduction in reconstruction error. For an image, the DCTHWT algorithm due to its real nature, is computationally simple and more accurate than the DFTHWT. Further compared to Cohen–Daubechies–Feauveau 9/7 biorthogonal symmetric wavelet, the DCTHWT, with its computational advantage, gives a better or comparable performance. For an image with 6.25% coefficients, the reconstructed image by DFTHWT is significantly inferior in appearance to that by DCTHWT which is reflected in the error index as its values are 3.0 and 2.65%, respectively. For spectral estimation, DCTHWT reduces the bias both in frequency (frequency resolution) and spectral magnitude. The reduction in magnitude bias in turn improves the signal detectability. In DCTHWT, the improvement in frequency resolution and the signal detectability is not only due to good quality DCT subband signals but also due to their stretching (decimation) in the wavelet transform. The MGD reduces the variance while preserving the frequency resolution achieved by DCT and decimation. In view of these, the new spectral estimator facilitates a significant improvement both in magnitude and frequency bias, variance and signal detection ability; compared to those of MGD processing of both DFT and DCT fullband and DFT subband signals.     

基于最小二乘支持向量机的衰落信道预测算法

该文探讨了利用相空间重构和支持向量机进行衰落信道非线性预测算法。该算法基于多径衰落信道具有混沌行为，利用坐标延迟理论，重建衰落信道系数的相空间，再根据混沌吸引子的稳定性和分形性，在相空间中通过递归最小二乘支持向量机（RLS-SVMM）进行预测。该算法对原始数据可以进行更平滑的处理，在噪声环境下预测的时间范围更长。对时间跨度为63．829ms的衰落系数进行了预测，仿真结果表明，在信噪比为15dB时，预测结果优于AR算法。     

基于自适应多尺度压缩感知的语音压缩与重构

本文针对语音信号的压缩感知问题,在系数总长度不超过原信号长度的前提下,推导了Sym小波分解合成的矩阵形式,提出了语音信号多尺度压缩感知(MCS)框架.进一步分析语音信号在小波基下不同级的稀疏性,提出了自适应多尺度压缩感知(AMCS)方法,把该方法运用到语音压缩与重构中,对重构语音进行了主客观评价,并进行了说话人识别验证...     

基于长短时能量均值的活动语音检测算法

为了有效抑制非平稳背景噪音对语音处理系统的严重干扰,提出了一种基于长短时能量均值的活动语音检测算法。该算法基于两个合理的假设,一个是基于语音隐含成分集的稀疏分解,不但能尽可能地深留含噪语音中的语音信息,还能在一定程度上消除非语音类噪音的干扰;另一个是对上述稀疏分解的语音进行重构,该重构信号中语音段的时域能量高于非语音段的时域能量。在上述两个假设的基础上,采用重构信号的时域能量作为音频特征,以当前帧为中心,并将与其相邻的特定数量帧的短时能量均值作为当前帧的得分值;以当前帧及其之前特定数量帧的长时能量均值怍为判决阈值,进而提出了以当前帧的短时能量均值和长时能量均值大小作为判断条件的活动语音检测算法。买验结果显示,该算法能有效地区分低信噪比（平稳噪音和忙平稳噪音）条件下的语音和非语音片段,并且其性能优于基于单Gaussian分布的似然比算法．     

基于压缩感知OMP改进算法的图像重构

正交匹配追踪(OMP)算法中迭代次数严格依赖信号的稀疏度K值,迭代次数选取适当会重构出高精确的图像,反之则会对图像重构质量造成严重影响.针对这一问题,提出了一种根据残差值的相对极差来确定最佳迭代次数的新方法.该方法要求在同一次迭代中对一幅图像的所有列同时进行迭代计算,根据极差的相对差值与门限值比较来确定最佳迭代次数,从而达到提高重构精度,消除对稀疏度K值依赖的目的.理论分析和仿真结果表明,改进的OMP算法比原有算法有更理想的重构效果,有更高的重构精度.     

Enhanced Speech Based Jointly Statistical Probability Distribution Function for Voice Activity Detection

Most of Voice activity detection (VAD) methods are based on statistical model. In these meth-ods, the noise signal is always assumed to satisfy and characterized by Gaussian distribution, while the assump-tion of noise does not always hold in practice and which causes that these kinds of method fail to distinguish speech from noise at low Signal-noise-ratio (SNR) level in non-stationary noise condition. For going further to improve the robustness of VAD, a enhanced speech based method is proposed. In the proposed method, the Laplacian distri-bution is used to model the remained noise since we find that the remained noise in enhanced speech satisfy Lapla-cian distribution; in addition, Gaussian mixture model is used to characterize the Discrete Fourier transform (DFT) coefficients of reconstructed speech in enhanced speech. Experimental results show that the proposed method per-forms better than the baseline method, especially in low SNR and non-stationary noise conditions.     

Time-varying parametric modeling of speech

For linear predictive coding (LPC) of speech, the speech waveform is modeled as the output of an all-pole filter. The waveform is divided into many short intervals (10–30 msec) during which the speech signal is assumed to be stationary. For each interval the constant coefficients of the all-pole filter are estimated by linear prediction by minimizing a squared prediction error criterion. This paper investigates a modification of LPC, called time-varying LPC, which can be used to analyze nonstationary speech signals. In this method, each coefficient of the all-pole filter is allowed to be time-varying by assuming it is a linear combination of a set of known time functions. The coefficients of the linear combination of functions are obtained by the same least squares error technique used by the LPC. Methods are developed for measuring and assessing the performance of time-varying LPC and results are given from the time-varying LPC analysis of both synthetic and real speech.