基于DFRCT滤波器组自适应语音增强算法

提出一种改进的语音增强方法,将带噪语音信号进行子带分解,再对子带信号进行离散分数余弦变换（DFRCT）域滤波,利用了DFRCT良好的正交特性,且自适应滤波采用最小均方（LMS）算法。对滤波后的信号进行DFRCT逆变换得到增强后的子带语音信号,合成增强后的语音信号。仿真结果表明,该算法在减少输入信号自相关程度的基础上,提高了收敛速度,减少了计算时间（约10 s）,增强后的语音信号的分段信噪比（SegSNR）和PESQ值都有所提高,具有良好的语音增强效果。     

基于子带无迹粒子滤波的语音增强算法

结合多采样率系统理论中的子带分解技术与贝叶斯估计理论中的无迹粒子滤波技术,提出了一种基于子带无迹粒子滤波的语音增强方法。该方法首先将语音信号分解成子带信号,建立各子带信号的低阶时变自回归模型;然后利用无迹粒子滤波估计模型参数,对子带信号进行滤波处理;最后根据滤波后的子带信号重构语音信号,实现语音增强。仿真结果表明,该方法能明显改善语音信号的信噪比和质量,且易于实现。     

滤波器组自适应语音增强方法

双麦克风噪声抵消应用中,由于交叉串的存在,传统自适应算法降噪性能受到很大的影响。为了提高双麦克风算法降噪性能,使用两级自适应滤波系统消除交叉串扰问题。为提高自适应滤波器收敛性能,采用主从结构LMS算法自适应调节步长因子。同时为了适合窄带处理算法,将输入信号进行子带分析预处理,对每个子带独立进行抗交叉串绕自适应处理,将各子带增强信号合并得到增强语音信号。实验结果表明,该方消噪量大,语音损伤小,语音增强效果显著。     

基于多统计模型和人耳听觉特性的麦克风阵列后滤波语音增强算法

针对麦克风阵列后滤波语音增强算法的不足, 结合人耳的听觉掩蔽效应, 提出了改进的后滤波语音增强算法. 提出了最大化目标语音存在概率来确定信号子空间维度的方法, 在噪声子空间上, 利用条件概率估计出噪声功率谱. 基于人耳的听觉掩蔽效应, 提出了后滤波器的一种合理的设计方法. 实验证明, 所提的噪声估计方法比传统方法更加准确, 所提的后滤波算法比传统的后滤波算法更好, 在多项语音评价指标上, 都取得了更好的实验效果.     

基于短时分数阶傅里叶变换的语音增强算法

提出了一种基于短时分数阶傅里叶变换(STFRFT)的语音增强新方法.该方法首先将带噪语音信号进行短时分数阶傅里叶变换,然后在分数阶傅里叶域(FRFD)对信号进行滤波,最后对滤波后的信号进行短时分数阶傅里叶逆变换,得到增强后的语音信号.实验表明在选定最佳的分数阶阶数时,可使噪声得到最大限度的滤除,大大提高了语音增强效果.     

EMD与自适应滤波相结合的语音增强法

为了提高语音信号的信噪比,提出一种经验模态分解与自适应滤波相结合的语音增强法。对带噪语音进行经验模态分解,得到有限个固有模态函数,把所有的固有模态函数按顺序分成三组,将每一组所包含的固有模态函数叠加,得到三个子信号;对三个子信号进行自适应滤波,消除噪声;将降噪后的子信号重构得到增强后的语音。仿真实验表明,所提方法的语音增强效果优于自适应滤波。     

基于子带分解的语音分离算法研究

提出一种有效解决不相互独立语音源信号混合的分离算法.利用子带分解方法,将混合信号分解成多个子带信号,在各个子带上分别进行语音分离得出语音分离信号,利用提出的相关性能指数,判断出相互独立的子带信号,把该子带的分离矩阵作为混合信号的解混合矩阵对混合信号进行分离.实验证明了本算法对相关语音源信号较好的分离效果.     

基于子带谱熵的仿生小波语音增强

前端噪声处理直接关系着语音识别的准确性和稳定性,针对小波去噪算法所分离出的信号不是原始信号的最佳估计,提出一种基于子带谱熵的仿生小波变换(BWT)去噪算法。充分利用子带谱熵端点检测的精确性,区分含噪语音部分和噪声部分,实时更新仿生小波变换中的阈值,精确地区分出噪声信号小波系数,达到语音增强目的。实验结果表明,提出的基于子带谱熵的仿生小波语音增强方法与维纳滤波方法相比,信噪比(SNR)平均提高约8%,所提方法对噪声环境下语音信号有显著的增强效果。     

基于听觉掩蔽效应的语音增强算法

对于低信噪比环境下的语音信号,传统谱减法残留的背景噪声较大。针对该问题,基于听觉掩蔽效应提出一种改进的语音增强算法。将人耳听觉掩蔽特性与功率谱减法相结合,设计一种时域递归平均算法对噪声进行估计,同时对带噪语音信号做频谱相减处理,从听觉的角度出发,利用估计的语音信号功率谱计算掩蔽阈值,并引入谱减功率修正系数和谱减噪声系数,实现带噪语音的信号增强。利用Matlab 2012b进行仿真,实验结果表明,该算法在低信噪比条件下能够较好地抑制背景噪声,改善语音质量,且与改进自适应滤波算法相比,其输出信号的信噪比可提高5%左右。     

基于TF-GSC的多通道后置滤波语音增强算法

针对传统语音增强算法在非平稳噪声,尤其是在噪声为语音的环境下,对噪声的抑制效果急剧下降的情况,提出了一种基于传递函数—广义旁瓣抵消(TF-GSC)和最佳修正测井谱振幅估计量(OM-LSA)的改进型多通道后置滤波语音增强算法.算法在后置滤波时,利用TF-GSC输出信号与参考噪声之间的相互关系求解出语音存在概率,并更新噪声功率谱估计.实验结果表明:算法可以有效地抑制非平稳噪声,提高语音增强算法在语音噪声环境下的鲁棒性.     

Performance of dual tone multi-frequency signal decoding algorithm using the sub-band non-uniform discrete Fourier transform on the ADSP-2192 processor

Efficient decoding of Dual Tone Multi-Frequency (DTMF) signals can be achieved using the sub-band non-uniform discrete Fourier transform (SB-NDFT). In this paper, the details of its implementation on the ADSP-2192 processor are put forward. The decoder performance in terms of its computational complexity and computational speed of this algorithm, implemented on the ADSP-2192 processor, are compared for different implementations of the SB-NDFT algorithm, with and without optimization for the chosen DSP, ADSP-2912. The algorithm is tested for various types of input signals on the DSP and these are compared with the results from Matlab^®. Problems on using other DTMF decoding algorithms that use the conventional discrete Fourier transform (DFT) and the non-uniform discrete Fourier transform (NDFT) are also addressed.     

基于FMFCC和HMM的说话人识别

美尔频率倒谱系数(MFCC)是说话人识别中常用的特征参数,而语音信号是非平稳信号,MFCC并不能很好的反映语音的时频特性。针对这一缺陷,为了提高说话人的识别率,结合新的时频分析工具分数傅立叶变换(FRFT)。将MFCC推广到分数形式,得到分数美尔频率倒谱系数(FMFCC),用以表征语音信号的特征;并利用可分性测度验证了特征参数的有效性;通过建立20个不同说话人的FMFCC特征库,采用隐马尔可夫模型(HMM)对说话人进行仿真识别。仿真结果表明,在合适的变换阶次下,说话人的平均识别率可达93%以上。     

基于Hermite特征矢量的二维离散分数维变换

分数维变换（FRFT）是分析时变信号的强有力工具之一,它可实现信号在时频域中任何角度的旋转。由于分数维变换的重要性,实现离散分数维变换（DFRFT）则显得很重要。根据DFRFT具有DFT的Hermite特征矢量这一特性,对二维离散信号进行二维DFRFT分析,此方法即可满足旋转特性,具又可获得与连续FRFT相似的结果。     

Improved voice activity detection algorithm using wavelet and support vector machine

This paper proposes an improved voice activity detection (VAD) algorithm using wavelet and support vector machine (SVM) for European Telecommunication Standards Institution (ETSI) adaptive multi-rate (AMR) narrow-band (NB) and wide-band (WB) speech codecs. First, based on the wavelet transform, the original IIR filter bank and pitch/tone detector are implemented, respectively, via the wavelet filter bank and the wavelet-based pitch/tone detection algorithm. The wavelet filter bank can divide input speech signal into several frequency bands so that the signal power level at each sub-band can be calculated. In addition, the background noise level can be estimated in each sub-band by using the wavelet de-noising method. The wavelet filter bank is also derived to detect correlated complex signals like music. Then the proposed algorithm can apply SVM to train an optimized non-linear VAD decision rule involving the sub-band power, noise level, pitch period, tone flag, and complex signals warning flag of input speech signals. By the use of the trained SVM, the proposed VAD algorithm can produce more accurate detection results. Various experimental results carried out from the Aurora speech database with different noise conditions show that the proposed algorithm gives considerable VAD performances superior to the AMR-NB VAD Options 1 and 2, and AMR-WB VAD.     

The fractional Fourier transform: theory, implementation and error analysis

The fractional Fourier transform is a time–frequency distribution and an extension of the classical Fourier transform. There are several known applications of the fractional Fourier transform in the areas of signal processing, especially in signal restoration and noise removal. This paper provides an introduction to the fractional Fourier transform and its applications. These applications demand the implementation of the discrete fractional Fourier transform on a digital signal processor (DSP). The details of the implementation of the discrete fractional Fourier transform on ADSP-2192 are provided. The effect of finite register length on implementation of discrete fractional Fourier transform matrix is discussed in some detail. This is followed by the details of the implementation and a theoretical model for the fixed-point errors involved in the implementation of this algorithm. It is hoped that this implementation and fixed-point error analysis will lead to a better understanding of the issues involved in finite register length implementation of the discrete fractional Fourier transform and will help the signal processing community make better use of the transform.     

Imperceptible digital watermarking scheme in multiple transform domains

A novel imperceptible digital watermarking scheme in multiple transform domains is presented, where the cover image is dealt with by discrete wavelet transform (DWT), discrete cosine transform (DCT) and discrete fractional random transform (DFRNT), while the watermark image is scrambled by Arnold transform and logistic map. First the watermark is scrambled by the Arnold transform, then the row and the column of the resulting watermark are scrambled by the Logistic map, respectively. In addition, four sub-band images are generated from the host image by the two-dimensional discrete wavelet transform. The low-frequency sub-band images are divided into 8?×?8 small matrices, and a coefficient matrix is produced by performing the discrete cosine transform on each matrix. An intermediate matrix with the same size as the watermark image is constructed by the intermediate frequency coefficients. Then the discrete fractional random transformation is performed on the intermediate frequency coefficient matrix and the scrambled watermark is embedded into the discrete fractional random transformation domain. Compared with the previous schemes, the proposed digital watermarking scheme has stronger imperceptibility and robustness.     

基于前置滤波和小波变换的带噪语音基音周期检测方法

根据语音信号的基音周期范围有限和在声门闭合时刻语音信号出现锐变的特点,提出一种基于前置滤波和小波变换的基音周期检测方法。带噪语音信号经过3阶椭圆低通滤波器滤波后,采用以二次样条小波作为小波函数,进行一级小波变换检测语音信号的锐变点,再计算基音周期。实验表明,本文提出的基音周期检测方法,与平均幅度差函数(AMDF)和自相关函数(ACF)方法相比,提高了提取基音周期的准确率;与多尺度小波变换的基音周期检测方法相比,减小了计算量,削弱了噪声信号和语音的共振峰对基音周期检测的影响。     

基于小波包变换的自适应门限的语音激活检测

语音激活检测是语音信号处理的一个重要环节.在低信噪比的情况下,传统的检测方法已不适用.为了提高语音激活检测的性能和鲁棒性,针对主要由白噪声组成的噪声背景,提出了一种基于小波包变换的自适应门限的语音激活检测方法(VAD),它将语音信号进行小波包变换,得到各个子带信号,符个子带信号通过Teager能量算子(TEO)将有声部分强化,同时衰减无声部分,最后进行自适应门限判决.实验结果表明在低信噪比的情况下,算法能够正确判别语音段和噪声段.