一种频域基频提取新方法

提出了一种基于二值侧抑制网络的频域基频提取方法。即利用二值侧抑制网络对语音的短时谱进行峰值提取,得到包括语音基频及其谐波的线谱,根据谐波间的距离平均值估算出基频．该算法在ＣＯＳＤＩＣ数据库上进行了测试,并与自相关基频提取法和倒谱基频提取法进行了比较．实验数据表明,本算法具有更高的精度和更强的抗噪声性能．     

长时语音特征在说话人识别技术上的应用

本文除介绍常用的说话人识别技术外,主要论述了一种基于长时时频特征的说话人识别方法,对输入的语音首先进行VAD处理,得到干净的语音后,对其提取基本时频特征。在每一语音单元内把基频、共振峰、谐波等时频特征的轨迹用Legendre多项式拟合的方法提取出主要的拟合参数,再利用HLDA的技术进行特征降维,用高斯混合模型的均值超向量表示每句话音时频特征的统计信息。在NIST06说话人1side-1side说话人测试集中,取得了18.7%的等错率,与传统的基于MFCC特征的说话人系统进行融合,等错率从4.9%下降到了4.6%,获得了6%的相对等错率下降。      

汉语普通话双基频检测

鉴于传统方法在语音双基频检测方面的局限性,本文提出了汉语双基频检测(DDPM)的方法.该方法利用混合汉语语音在短时帧之内的准周期性,经搜索得到每一帧内的双基频候选点,再根据正常情况下人的发音在相邻帧内基频不产生突变的特点,实现了双基频检测.应用此方法,在纯净与加噪的情形下,对汉语四种声调单音节的各种组合分别进行了检测实验,检测效果十分理想.新方法也可以直接应用到连续语音双基频的检测.     

融合梅尔谱增强与特征解耦的噪声鲁棒语音转换

提出了一种融合梅尔谱增强与特征解耦的噪声鲁棒语音转换模型,即MENR-VC模型。该模型采用3个编码器提取语音内容、基频和说话人身份矢量特征,并引入互信息作为相关性度量指标,通过最小化互信息进行矢量特征解耦,实现对说话人身份的转换。为了改善含噪语音的频谱质量,模型使用深度复数循环卷积网络对含噪梅尔谱进行增强,并将其作为说话人编码器的输入;同时,在训练过程中,引入梅尔谱增强损失函数对模型整体损失函数进行了改进。仿真实验结果表明,与同类最优的噪声鲁棒语音转换方法相比,所提模型得到的转换语音在语音自然度和说话人相似度的平均意见得分方面,分别提高了0.12和0.07。解决了语音转换模型在使用含噪语音进行训练时,会导致深度神经网络训练过程难以收敛,转换语音质量大幅下降的问题。     

一种定征复合板材粘接层性质的非线性超声兰姆波方法

借助于兰姆波频散曲线及导波激发的模式展开分析方法,对基频兰姆波时域信号及二次谐波时域信号的发生过程进行了直观的论述。结合Ritec-SNAP系统的测量功能,详细分析了二次谐波时域脉冲包络积分表达式的物理意义;该积分表达式可表征基频兰姆波时域脉冲传播过程中的二次谐波发生效率,以及基频与二倍频兰姆波模式之间的频散程度。在基频与二倍频兰姆波相速度相等(或近似相等)的频率附近,实验观察到显著的且无模式混叠的二次谐波信号,显示出在兰姆波的传播过程中的确可存在强烈的非线性效应。对于三种不同粘接情形的复合板材,实验结果表明,采用本文引入的非线性兰姆波应力波因子,结合二次谐波幅频曲线峰值所对应的频率值,可有效地对板材粘接层性质进行表征。     

声门下共鸣的谱规整用于非特定人的语音识别

提出在参数的提取过程中用不同的感知规整因子对不同人的参数归一化,从而实现在非特定人语音识别中对不同人的归一化处理。感知规整因子是基于声门上和声门下之间耦合作用产生声门下共鸣频率来估算的,与采用声道第三共振峰作为基准频率的方法比较,它能较多的滤除语义信息的影响,更好地体现说话人的个性特征。本文提取抗噪性能优于Mel倒谱参数的感知最小方差无失真参数作为识别特征,语音模型用经典的隐马尔可夫模型(HMM)。实验证明,本文方法与传统的语音识别参数和用声道第三共振峰进行谱规整的方法相比,在干净语音中单词错误识别率分别下降了4%和3%,在噪声环境下分别下降了9%和5%,有效地改善了非特定人语音识别系统的性能。      

采用低维特征映射的耳语音向正常音转换

在将耳语音转换为正常音时,为了研究降维后语音特征对耳语音转换的影响,分别对耳语音和正常音谱包络进行自适应编码以提取耳语音和正常音的低维特征,然后使用BP网络建立耳语音和正常音低维谱包络特征之间的映射关系以及正常音基频和耳语音低维谱包络特征之间的关系。转换时,根据耳语音低维谱包络特征获得对应正常音的低维谱包络特征和基频,对低维谱包络特征进行解码后获得对应的正常音谱包络。实验结果表明,采用此方法转换后的语音与正常音之间的倒谱距离相比高斯混合模型方法下降了10%,转换后语音的自然度和可懂度都有所提高。      

兴奋模式下谐波复合音音高感知机制的研究

通过测量谐波复合音的基频辨别阈,探讨中等"高次谐波"的音高感知是否依赖于谐波的可分离性,以及掩蔽音对实验结果的影响。实验方法:在目标音单独存在或目标音与掩蔽音混合时,将刺激通过高、中、低三个带通滤波器以获得不同的谐波可分离度。实验刺激设计为5种基频差异和4种相位组合。五名被试均为年轻人,纯音听阈≤15 dB HL。研究结果发现:谐波复合音的基频辨别阈随着信号频段的上移而增大;目标音和掩蔽音的基频差异对基频辨别阈有显著影响;但相位影响不显著。结论:谐波的可分离性对基频辨别阈有显著影响,但中等"高次谐波"的音高感知不依赖于可分离性;混合音的大部分音高感知结果与兴奋模式的峰值大小密切相关。      

改进LVAMDF及综合多因素基音检测算法

在对语音信号静音、清音、浊音划分的基础上,针对语音信号周期特征明显段分布随机性问题,提出改进的变长度平均幅度差函数LVAMDF及综合多因素基音检测算法,该算法对语音信号进行周期特征明显段和周期特征不明显段的聚类划分,同时,获取周期特征明显语音段的基音周期,针对少数基音周期划分倍频或半频问题,提出识别、修正方法,其识别、修正率极高。在对大量真实语音处理中,能够精确的检测出语音特征明显段的基音周期端点,基本没有倍频和半频划分,并且和AMDF、ACF算法作了对比。     

基于小波变换的重叠语音基频提取及声调识别

提出一种基于小波变换的重叠语音基频提取及声调识别的方法。利用小波的伸缩和时移特性，通过对重叠语音做多尺度的小波变换，可以有效地提取重叠语音中各自的基音频率，并在此基础上实现声调的识别。实验表明，此方法是有效的，是重叠语音基频提取及声调识别的一种新途径。     

矩形辐射器压电多频超声换能器的研究

研究了一种具有多个共振频率的矩形辐射器夹心式超声换能器,换能器由圆柱形后盖板、压电陶瓷晶堆及矩形六面体辐射器前盖板组合而成。利用表观弹性法和一维近似理论给出了多频换能器横向及纵向理论共振频率方程。对一种特殊情况下的此类换能器进行了有限元及实验分析,给出了各自的频率输入导纳曲线。对理论和实验结果进行分析后表明,此类矩形辐射器夹心式超声换能器可以在不同的振动模态上工作,具有多个共振频率.     

In situ FH analysis at disk drive level

Drive-level FH evaluation is important in terms of FH control, FH adjustment and disk drive robustness inside operating hard disk drives. Characterization of FH requires simple methodology, easy implementation in addition to the general requirement to in situ FH analysis. This paper reports authors’ effort in proposing a new FH error function to evaluate harmonic ratio FH methods; and based on the error function, a new FH measurement method named multi-frequency method, which calculates FH by reading back multi-frequency pattern, is proposed to minimize FH measurement error by optimizing pattern frequencies. Such technology is also applied at disk drive level to investigate FH variation when a hard disk drive is operating under variable environmental conditions.     

Nonlinear multi-frequency generation for underwater application

This study examines the feasibility of designing a multi-frequency acoustic surveying tool based on the saturation effect. The transmitter is driven by a high-power single tone-burst: nonlinear propagation creates the beams at harmonic frequencies. A simple pseudo one-dimensional model is used to estimate the expected on-axis harmonic levels generated with a rectangular aperture. First measurements are reported and compared with the estimations.     

Multi-frequency inverse-phase fringe projection profilometry for nonlinear phase error compensation

A new multi-frequency inverse-phase method was proposed to compensate for nonlinear phase errors in fringe projection profilometry and to measure the three-dimensional shape of discontinuous objects. After introducing a phase offset of π/4 into the multi-frequency four-step phase-shifting method the corresponding nonlinear phase error reversed its sign, which allowed the addition of unwrapped phases before and after the phase-offset operation to compensate for the error. For the four-step phase-shifting method, simulation analysis showed that the nonlinear phase error had quadrupled the fringe frequency. Moreover, experimental results verified the feasibility and applicability of the proposed method.     

Research on stimulated Brillouin scattering suppression based on multi-frequency phase modulation

The method of generating equal-amplitude spectral lines by multi-frequency phase modulation is used in stimulated Brillouin scattering （SBS） suppression. The spectra of three, five, seven, and eleven equalamplitude spectral lines are obtained in experiment with flatnesses less than 0.3 dB. Theoretical research on SBS suppression shows that the threshold power after modulation is in reverse proportion to the maximum square of amplitude moduli of fundamental frequency and the nth harmonic wave. The threshold powers of three, five, seven, and eleven equal-amplitude spectral lines are improved by 5.21, 8.36, 9.39, and 10.76 dB, respectively.     

稀磁半导体(Ga,Mn)As薄膜激光诱导超快磁化动力学过程拟合方法探究

本文对稀磁半导体(Ga, Mn)As薄膜中超快激光诱导磁化动力学响应信号的不同拟合方法进行了对比分析. 通过Landau-Lifshitz-Gilbert(LLG)方程的数值拟合发现, 由于薄膜平面内和平面外磁光响应强度不同, 磁矢量三维进动的叠加可以导致多个频率振动模式的假象. 当使用高于(Ga, Mn)As带边的能量激发时, 磁化进动的磁光响应信号中叠加着来自光极化载流子的响应, 此时单纯利用LLG方程对薄膜整体磁化动力学过程拟合应谨慎使用. 本工作为正确分析和理解脉冲激光对(Ga, Mn)As铁磁性的超快调控提供了拟合方法上的指导.     

Optimum projection pattern generation for grey-level coded structured light illumination systems

Structured light illumination (SLI) systems are well-established optical inspection techniques for noncontact 3D surface measurements. A common technique is multi-frequency sinusoidal SLI that obtains the phase map at various fringe periods in order to estimate the absolute phase, and hence, the 3D surface information. Nevertheless, multi-frequency SLI systems employ multiple measurement planes (e.g. four phase shifted frames) to obtain the phase at a given fringe period. It is therefore an age old challenge to obtain the absolute surface information using fewer measurement frames. Grey level (GL) coding techniques have been developed as an attempt to reduce the number of planes needed, because a spatio-temporal GL sequence employing p discrete grey-levels and m frames has the potential to unwrap up to p^m fringes. Nevertheless, one major disadvantage of GL based SLI techniques is that there are often errors near the border of each stripe, because an ideal stepwise intensity change cannot be measured. If the step-change in intensity is a single discrete grey-level unit, this problem can usually be overcome by applying an appropriate threshold. However, severe errors occur if the intensity change at the border of the stripe exceeds several discrete grey-level units. In this work, an optimum GL based technique is presented that generates a series of projection patterns with a minimal gradient in the intensity. It is shown that when using this technique, the errors near the border of the stripes can be significantly reduced. This improvement is achieved with the choice generated patterns, and does not involve additional hardware or special post-processing techniques. The performance of that method is validated using both simulations and experiments. The reported technique is generic, works with an arbitrary number of frames, and can employ an arbitrary number of grey-levels.     

Error-compensating algorithms in phase-shifting interferometry: a comparison by error analysis

In this paper, two families of phase-shifting algorithms with π/2 phase steps are studied. In family I, three new algorithms are derived by using the averaging technique based on the Surrel six-sample algorithm with phase shifts of π/2. Family II includes four well-known algorithms derived by the averaging technique based on the conventional four-sample algorithm with π/2 phase steps. A polynomial model of phase-shift errors used to describe general expressions for calculation of the correct object phase via the Fourier spectra analysing method as a function of the harmonic order in the fringe signal is presented. The error-compensating properties of the algorithms in families I and II are investigated by the Fourier spectra analysing method. It is found that the averaging technique, when used in any of the algorithm with π/2 phase steps, can improve the phase-shifting algorithm property: it is insensitive to phase-shift error when the fringe signal contains the first harmonic, but it can't be used to enhance the phase-shifting algorithm properties when the fringe signal contains higher order harmonics (n2). P–V (peak–valley) phase errors are calculated by the computer simulation and tables and plots are presented, from which the algorithms in families I and II are compared. It is shown that the algorithms in family I are more insensitive to phase-shift errors when the fringe signal contains the second harmonic and the algorithms in family II are more insensitive to phase-shift errors when the fringe signal is a sinusoidal waveform.     

SFC轴向注入系统和中心区的研究和设计

介绍了在HIRFL注入器SFC上进行的中心区、聚束器系统和轴向注入束运线的设计、加工和调试结果.中心区的设计采用了两种注入半径及相应的两套螺旋式静电偏转镜,解决了高频电压在某些工作区域偏低及三次谐波加速时轴向注入线上空间电荷效应较为严重的问题.新的锯齿波聚束器系统不仅可以提高聚束效率,而且还提出了采用半频聚束模式以提高SFC与主加速器SSC?的纵向匹配效率.新设计的轴向注入束运线配备了两台在线ECR离子源,提高了电荷态分辨能力和注入相空间匹配能力,在提高注入效率的同时还改善了离子源及束运线的工作环境和调束手段.     

The norms and variances of the Gabor, Morlet and general harmonic wavelet functions

This paper deals with certain properties of the continuous wavelet transform and wavelet functions. The norms and the spreads in time and frequency of the common Gabor and Morlet wavelet functions are presented. It is shown that the norm of the Morlet wavelet function does not satisfy the normalization condition and that the normalized Morlet wavelet function is identical to the Gabor wavelet function with the parameter σ=1.The general harmonic wavelet function is developed using frequency modulation of the Hanning and Hamming window functions. Several properties of the general harmonic wavelet function are also presented and compared to the Gabor wavelet function. The time and frequency spreads of the general harmonic wavelet function are only slightly higher than the time and frequency spreads of the Gabor wavelet function. However, the general harmonic wavelet function is simpler to use than the Gabor wavelet function. In addition, the general harmonic wavelet function can be constructed in such a way that the zero average condition is truly satisfied. The average value of the Gabor wavelet function can approach a value of zero but it cannot reach it.When calculating the continuous wavelet transform, errors occur at the start- and the end-time indexes. This is called the edge effect and is caused by the fact that the wavelet transform is calculated from a signal of finite length. In this paper, we propose a method that uses signal mirroring to reduce the errors caused by the edge effect. The success of the proposed method is demonstrated by using a simulated signal.