按规则合成无限词汇汉语语声的研究

本文研究了由计算机输入汉语拼音字符,以汉语音素作为声元素,根据语音的韵律规则合成连续汉语语声的方法。该方法数据量少,便于控制语音、语调。实验表明,利用该方法构成汉语文-语转换系统是可行的。     

汉语文语转换系统的语音规则和声学参数

利用瑞典皇家理工学院（KTH）语言通讯和音乐声学系Carlson和Granstrom设计的多语种文语转换系统,进行了汉语文语转换研究。对汉语文语转换用的基本语音规则和声学参数做了分析、设计和调试,特别对轻声、儿化和变调进行了研究。     

粤语文语转换系统研究

由于计算机辅助数学(CAI)发展的需要,我们基于IBM系列微机,建立了一个粤语(广州话)文语转换系统。通过对粤语发音规律及语音参数的研究,此系统应用了共振峰参数语音合成技术,以无调音节和声调模式为基本合成单元,从自然语流中提取每个音节的共振峰语音参数并建立了供合成用的语音参数库。为实现计算机上的文语转换,自行编制了语音管理程序。此系统实现了国家一、二级汉字的广州话读音。日、英、俄文字母的发音及若干科学符号的广州话读音。且实现了计算机显示及相应发音的自由控制。     

A text-to-speech system with high intelligibility and naturalness for Chinese

I.IntroductionResearchesonChinesesynthesisdisclosethatonlywhenboththesegmentalandsupraseg-melltalfeaturesofthesyntheticspeecharesimilartothoseofthellaturalone,thesyntheticspeechwillsoundintelligibleandnatural[1].Amongekistingsynthetictechniques,theapproachbasedonacousticparametersca-nadustboththesegmentalandsuprasegmentalfeaturesofsyntheticunitsfiekiblyandcanbeconsideredasthemostreasonablesynthetictechniqueintheory.However,theparameterbasedsynthesizerisoverAfependentonthedevelopmentsofparamet…     

Use of the magnitude estimation technique for assessing the performance of text-to-speech synthesis systems

As text-to-speech systems develop, it becomes necessary to compare various solutions and to evaluate whether a change in the synthesis procedure has an effect on the listener's attitude to the system. The possibility of directly scaling intelligibility, naturalness, and user's satisfaction (i.e., acceptability) with the magnitude estimation technique is investigated. A magnitude estimation protocol suitable for this purpose is described. In general, within the limits of the methodological constraints discussed in this paper, the procedure appears to be reliable and valid for quantifying the perceived attributes of synthesized speech.     

一种高清晰度、高自然度的汉语文语转换系统

以基音同步叠加技术为基础,以汉语单音节为合成单元,有一包含词调模式、重音模式和句调模式的韵律规则库的汉语文语转换系统,可合成出高清晰度和高自然度的汉语语音。研究表明,影响汉语合成语音的自然度的主要因素是音高和音强随时间的变化、各音节的音长分布以及音节间的协同发音,其中以音高和音长的影响最为显著。时域基音同步叠加技术提供了一种在时域改变语音波形的音高和音长的方法,从而使在用波形拼接法合成汉语时,进行词一级和句一级的韵律调节成为可能。对新闻广播语言的声学特征的分析,为建立汉语合成的韵律调节规则提供了理论依据。本文介绍新的汉语文语转换系统的结构及流程、对广播语言韵律特征的初步研究结果、汉语合成规则及合成系统语音质量的评测结果。     

基于统计声学模型的单元挑选语音合成算法

在多语种语音合成任务中, 由于单人多语种数据稀缺, 让一个音色同时支持多种语言合成变得非常困难。不同于已有方法仅在声学模型中解耦音色和发音, 提出一种融合跨说话人韵律迁移的端到端多语种语音合成方法, 采用两级层级条件变分自编码器直接建模从文本到波形的生成过程, 并解耦了音色、发音和韵律等信息。该方法通过迁移目标语种已有说话人的韵律风格来改善跨语种合成的韵律。实验表明, 所提模型在跨语种语音生成上获得了3.91和4.01的自然度和相似度平均意见得分, 相比基线跨语种合成字错误率降低到5.85%。韵律迁移以及消融实验也进一步证明了该方法的有效性。

     

面向情感语音转换的韵律转换方法

面向情感语音转换,该文提出了一种韵律转换方法。该方法包含基频转换和时长转换两个部分,前者选择离散余弦变换(DCT)参数化基频,根据基频的层次结构特点,将基频分解为短语层和音节层两个层次,使用基于混合高斯模型(GMM)的转换方法对两个层次分别进行转换;后者使用基于分类回归树(CART)的方法以声韵母为基本单位对时长进行转换。一个包含三种基本情感的语料库用作训练和测试,客观评测以及主观评测实验结果显示该方法可有效进行情感韵律转换,其中悲伤情感在主观实验中达到了接近100%的正确率。     

Nanotaper for compact mode conversion

We propose and demonstrate an efficient coupler for compact mode conversion between a fiber and a submicrometer waveguide. The coupler is composed of high-index-contrast materials and is based on a short taper with a nanometer-sized tip. We show that the micrometer-long silicon-on-insulator-based nanotaper coupler is able to efficiently convert both the mode field profile and the effective index, with a total length as short as 40 microm. We measure an enhancement of the coupling efficiency between an optical fiber and a waveguide by 1 order of magnitude due to the coupler.     

Double-rod metasurface for mid-infrared polarization conversion

Resonant responses of metasurface enable effective control over the polarization properties of lights. In this paper,we demonstrate a double-rod metasurface for broadband polarization conversion in the mid-infrared region. The metasurface consists of a metallic double-rod array separated from a reflecting ground plane by a film of zinc selenide. By superimposing three localized resonances, cross polarization conversion is achieved over a bandwidth of 16.9 THz around the central frequency at 34.6 THz with conversion efficiency exceeding 70%. The polarization conversion performance is in qualitative agreement with simulation. The surface current distributions and electric field profiles of the resonant modes are discussed to analyze the underlying physical mechanism. Our demonstrated broadband polarization conversion has potential applications in the area of mid-infrared spectroscopy, communication, and sensing.     

一种面向声音变换的参数化模型

在源滤波器模型的基础上,利用统计学习方法,建立了一种面向声音变换的混合参数化模型。该模型包括浊音声学模型、清音声学模型和韵律补偿模型三部分。基于线性预测分析和mel倒谱分析的浊音声学模型,刻画了说话人声腔的共振特性。基于线性预测分析和噪声源分析的清音声学模型,反映了说话人发清音的特点。基于统计学习方法的韵律补偿模型描述了音高、能量与时长等分布特性。在该混合参数化模型的基础上,提出了一个声音变换算法,并将其应用到汉语音节的变换问题上。实验结果表明,对清浊音和韵律特性分别建模的变换算法能够提高重建语音的清晰度和可懂度,缩小重建语音与目标语音之间的感知距离,使重建语音具有目标说话人的韵律特征.     

Mechanism for structure conversion of FeSi under pressures

The atomistic mechanism for the structure conversion of FeSi from B20 to B2 phase at high pressures is a long-standing problem in Fe-Si system. Here we explore the atomistic mechanism about this intriguing structure conversion by first-principle calculations. We find the structure conversion from B20 to B2 is inhibited by the high kinetic barriers and the weak driving forces from low relative enthalpy under pressures. We also identify an orthorhombic structure with Pbcm symmetry as an intermediate meta-stable phase along the minimum energy path, which is possibly the precursory state of the B2 phase. These results can help to explain experimental findings and explore the kinetic origin of this intriguing structure conversion of FeSi at high pressure and temperature.     

Terahertz generation based on parametric conversion: from saturation of conversion efficiency to back conversion

By stacking alternatively rotated gallium phosphide (GaP) plates, the maximum photon conversion efficiency of 40% for the terahertz (THz) generation based on difference-frequency generation has been achieved. The corresponding peak power generated inside the four GaP plates approaches 4 kW. As the number of plates is increased from four to five, the THz output power is significantly decreased, due to back parametric conversion.     

Picosecond bismuth-doped fiber MOPFA for frequency conversion

We report the development of a bismuth-doped fiber master oscillator power fiber amplifier system. The system operates at 1177 nm, producing 28 ps pulses at 9.11 MHz repetition rate, with an output power of 150 mW and a peak pulse power of 580 W. We subsequently frequency double the output, resulting in a picosecond pulsed visible source operating at 588.5 nm, with a maximum average output power of 13.7 mW.     

A ray-based algorithm for multi-dimensional linear conversion

A numerical algorithm is proposed for connecting the incoming and outgoing wave fields in studies of linear conversion. This is the first such ray-based algorithm for wave conversion in multiple spatial dimensions. It is demonstrated that, aside from the overall phase of the coupling, one can directly evaluate all quantities needed for the connection coefficients from the ray geometry. The ray dynamics is generated using the determinant of the dispersion matrix as the Hamiltonian. Using information available while following an incoming ray, the algorithm automatically detects that the ray has entered a conversion region, evaluates the transmission and conversion coefficients, and launches the transmitted ray. The algorithm does not require any prior knowledge of the geometry of the conversion region. The algorithm is illustrated using a two-dimensional toroidal model with resonant conversion from a magnetosonic to an ion-hybrid wave.     

Photonic quasicrystals for nonlinear optical frequency conversion

We present a general method for the design of 2-dimensional nonlinear photonic quasicrystals that can be utilized for the simultaneous phase matching of arbitrary optical frequency-conversion processes. The proposed scheme--based on the generalized dual-grid method that is used for constructing tiling models of quasicrystals--gives complete design flexibility, removing any constraints imposed by previous approaches. As an example we demonstrate the design of a color fan--a nonlinear photonic quasicrystal whose input is a single wave at frequency omega and whose output consists of the second, third, and fourth harmonics of omega, each in a different spatial direction.