基于并行基因算法的语音识别方法

提出一种基于并行基因算法的孤立字识别时间规正算法，该算法是在［３］的基础上提出，可解决动态时间规划（ＤＴＷ）难以解决的一些问题：①使距离归一化因子Ｍ与实际路径相关；②以自然方式提供多条最佳规划路径；③语音端点检测正确性对识别率的影响得到一定程度的改善。建立了试验数据库，根据试验数据建立了模板距离遵循正态分布的算法性能分析模型。比较了并行基因算法，串行基因算法［３］和动态时间规划算法的性能。试验结果表明：基因算法比动态时间规划能得到更高的识别率，在单ＣＰＵ情形下，虽然并行基因算法的性能比串行基因算法略微提高，但至少可节约三分之一的ＣＰＵ时间     

子带特征存英语词重音检测中的应用

对于英语等"重音节拍语言",词重音是一个非常重要的韵律学特征、提出巴克谱子带能量(BSSE),巴克谱子带倒谱(BSSC)和巴克倒谱(BSCC)等三种使用子带方法,模拟人类听觉系统进行英语词重音检测的新特征.首先研究了在英语词重音检测中不同频带的贡献度,然后测试了在连续语音条件下使用这些特征的系统性能.试验结果表明,高频带对重音识别的贡献度比其它频带大,新特征的识别效果也比传统特征好.     

Structural definition of a conserved neutralization epitope on HIV-1 gp120

The remarkable diversity, glycosylation and conformational flexibility of the human immunodeficiency virus type 1 (HIV-1) envelope (Env), including substantial rearrangement of the gp120 glycoprotein upon binding the CD4 receptor, allow it to evade antibody-mediated neutralization. Despite this complexity, the HIV-1 Env must retain conserved determinants that mediate CD4 binding. To evaluate how these determinants might provide opportunities for antibody recognition, we created variants of gp120 stabilized in the CD4-bound state, assessed binding of CD4 and of receptor-binding-site antibodies, and determined the structure at 2.3 A resolution of the broadly neutralizing antibody b12 in complex with gp120. b12 binds to a conformationally invariant surface that overlaps a distinct subset of the CD4-binding site. This surface is involved in the metastable attachment of CD4, before the gp120 rearrangement required for stable engagement. A site of vulnerability, related to a functional requirement for efficient association with CD4, can therefore be targeted by antibody to neutralize HIV-1.     

Antibody neutralization and escape by HIV-1

Neutralizing antibodies (Nab) are a principal component of an effective human immune response to many pathogens, yet their role in HIV-1 infection is unclear. To gain a better understanding of this role, we examined plasma from patients with acute HIV infection. Here we report the detection of autologous Nab as early as 52 days after detection of HIV-specific antibodies. The viral inhibitory activity of Nab resulted in complete replacement of neutralization-sensitive virus by successive populations of resistant virus. Escape virus contained mutations in the env gene that were unexpectedly sparse, did not map generally to known neutralization epitopes, and involved primarily changes in N-linked glycosylation. This pattern of escape, and the exceptional density of HIV-1 envelope glycosylation generally, led us to postulate an evolving 'glycan shield' mechanism of neutralization escape whereby selected changes in glycan packing prevent Nab binding but not receptor binding. Direct support for this model was obtained by mutational substitution showing that Nab-selected alterations in glycosylation conferred escape from both autologous antibody and epitope-specific monoclonal antibodies. The evolving glycan shield thus represents a new mechanism contributing to HIV-1 persistence in the face of an evolving antibody repertoire.     

Recent Advances in the Development of Polymer-supported Phosphines

1 Introduction Recent years have seen a growing use of polymer-supported reagents by organic chemists in traditional solution-phase synthesis.Of the many reagents supported by polymers available, variations of triphenylphosphine are among the most broadly used. This is because triphenylphosphine is not only a reagent in a wide range of organic reactions~([1]), but it also serves as a ligand in many organometallic reagents~([2]).Most importantly, polymer-supported triphenylphosphine affords the…     

一种改进的BIC说话人改变检测算法

针对贝叶斯信息准则(BIC)算法在说话人改变检测中计算量大、检测精度低的问题,文中提出了一种改进的BIC说话人改变检测算法.该算法通过限制分析窗内第一个数据窗的最大长度来降低计算量,并通过增加分析窗内第二个数据窗的有效长度(提高可测度)来提高检测精度;同时,该算法只在新增区间内寻找潜在说话人改变点,从而解决了长时间无说话人改变时计算量不断增大的问题.实验结果表明,该算法和传统的BIC算法相比,偏移误差范围由0.10～0.80降低到0.03～0.20;当分析窗长为40s时,计算时间节省了约75%.     

基于唇重构与三维耦合CNN的多视角音唇一致性判别

针对传统音唇一致性判别方法主要对正面唇动视频进行处理,未考虑视频采集角度变化对结果的影响,且容易忽略唇动过程中的时空特性等不足,文中以唇部角度变化对一致性判别的影响为研究重心,结合三维卷积神经网络在非线性表示和时空维度特征提取上的优势,提出了基于正面唇重构与三维耦合卷积神经网络的多视角音唇一致性判别方法。该方法先通过在生成器中引入自映射损失来提高正面重建效果,并采用基于自映射监督循环一致性生成对抗网络（SMS-CycleGAN）的唇重构方法对多视角唇图进行角度分类及正面重构;然后设计两个异构三维卷积神经网络,分别用来描述音频和视频信号,并提取包含长时时空关联信息的三维卷积特征;最后引入对比损失函数作为音视频信号匹配的相关度鉴别度量,将音视频网络输出耦合到同一表示空间,并进行一致性判别。实验结果表明,文中方法能重建出更高质量的正面唇图,一致性判别性能优于多种不同类型的比较方法。     

可证明安全的基于口令的动态认证方案

在挑战/响应方案的基础上,结合安全单向Hash函数提出了一种基于口令的、安全的动态认证方案,用户不需改变使用口令认证的习惯,而认证方案比传统的基于口令的方案更为安全。通过Random Oracle模型分析,证明了只要选取安全的单向Hash函数,现提出的基于口令的动态认证方案是安全的。     

生物反应器内人参细胞的培养生长及代谢特性

利用2．4L搅拌式反应器对人参细胞的培养生长与代谢特性进行了研究．结果显示：细胞3个生长时期的持续时间与接种量、初始营养浓度、营养流加情况及培养条件（如温度和溶解氧等）有关；分批培养和流加培养时的最大细胞得率分别为12g/L和35g/L（细胞干重）；细胞的皂甙含量在旺盛生长期快速上升，并保持增长至衰退期前期；干细胞中最大的皂甙含量达60～70g／kg；随着细胞的生长，培养液的电导值从8000μS以上下降至600μS以下，细胞自溶后逐渐回升．另外，培养后期部分细胞颜色变红，但没有发现颜色转变对皂甙含量造成影响．培养液的电导可以作为判别培养状态的指标之一，用于指导培养控制及终止培养操作．细胞的收获时间应在衰退期的前期．