基于多流多状态动态贝叶斯网络的音视频连续语音识别

语音和唇部运动的异步性是多模态融合语音识别的关键问题,该文首先引入一个多流异步动态贝叶斯网络(MS-ADBN)模型,在词的级别上描述了音频流和视频流的异步性,音视频流都采用了词-音素的层次结构.而多流多状态异步DBN(MM-ADBN)模型是MS-ADBN模型的扩展,音视频流都采用了词-音素-状态的层次结构.本质上,MS-ADBN是一个整词模型,而MM-ADBN模型是一个音素模型,适用于大词汇量连续语音识别.实验结果表明：基于连续音视频数据库,在纯净语音环境下,MM-ADBN比MS-ADBN模型和多流HMM识别率分别提高35.91%和9.97%.     

基于多流动态贝叶斯网络的音视频连续语音识别

针对说话时发音和口形的异步问题,提出了一个多流异步动态贝叶斯网络(DynamicBayesian Network,DBN)模型,以实现基于音视频特征的连续语音识别,在这个模型中,音频流和视频流在词节点同步,而在词节点之间,音视频流有各自独立的拓扑结构以及节点变量之间的条件依赖关系,同时词转移节点变量由音视频流共同确定,模型在词级别上体现了音视频流的异步性.采用连续数字音视频数据库的实验结果表明,在信噪比为O～30 dB的测试环境下,比较单流DBN模型和多流隐马尔可夫模型,平均识别率分别提高了8.68%和10.07%.     

Development of a thermogelling ophthalmic formulation of cysteine

Preliminary studies carried out with cysteine 2% solution showed that pH adjusted to isoelectrical pH (i.e., 4.9) led to enhance stability during autoclaving and ensured no significant degradation during at least 14 days if stored at 2-8°C protected from light. Optimized formulations combined either cysteine(2%)/Poloxamer407(16.5%) or cysteine(2%)/Poloxamer407(20%)/Poloxamer188(5%) and were characterized by an adequate temperature of gelification (TG) (25.9°C and 26.9°C, respectively), an important gel strength (5.1daN and 5.3daN, respectively) and a drastic increase in the apparent viscosity between 24°C and 32°C (multiplication factor of 78 and 77-fold, respectively). Cysteine addition produced only slight but significant decrease in temperature of gelification and increase in gel strength.     

基于谐波的乐纹提取和音乐检索

提出了一种基于节拍内音乐谐波特性的乐纹特征提取方法,首先求得每帧音乐的谐波信息,再利用跟踪得到的节拍,计算节拍内所有帧的谐波信息的均值,构成此节拍的乐纹特征矩阵。为了提高音乐检索的效率,设计了一个二级音乐检索算法：根据节拍信息,将与查询音乐片段的每分钟节拍数相近的音乐作为候选音乐,再逐节拍计算所查询音乐的乐纹和候选音乐乐纹的相似度,选择相似度最高的音乐作为检索结果。实验结果表明,提出的乐纹特征和音乐检索算法有效地提高了检索准确率和检索效率。     

A hierarchical Markovian model for multiscale region-based classification of vector-valued images

We propose a new classification method for vector-valued images, based on: 1) a causal Markovian model, defined on the hierarchy of a multiscale region adjacency tree (MRAT), and 2) a set of nonparametric dissimilarity measures that express the data likelihoods. The image classification is treated as a hierarchical labeling of the MRAT, using a finite set of interpretation labels (e.g., land cover classes). This is accomplished via a noniterative estimation of the modes of posterior marginals (MPM), inspired from existing approaches for Bayesian inference on the quadtree. The paper describes the main principles of our method and illustrates classification results on a set of artificial and remote sensing images, together with qualitative and quantitative comparisons with a variety of pixel-based techniques that follow the Bayesian-Markovian framework either on hierarchical structures or the original image lattice.     

A video prediction approach for animating single face image

Generating dynamic 2D image-based facial expressions is a challenging task for facial animation. Much research work focused on performance-driven facial animation from given videos or images of a target face, while animating a single face image driven by emotion labels is a less explored problem. In this work, we treat the task of animating single face image from emotion labels as a conditional video prediction problem, and propose a novel framework by combining factored conditional restricted boltzmann machines (FCRBM) and reconstruction contractive auto-encoder (RCAE). A modified RCAE with an associated efficient training strategy is used to extract low dimensional features and reconstruct face images. FCRBM is used as animator to predict facial expression sequence in the feature space given discrete emotion labels and a frontal neutral face image as input. Both quantitative and qualitative evaluations on two facial expression databases, and comparison to state-of-the-art showed the effectiveness of our proposed framework for animating frontal neutral face image from given emotion labels.

     

Development of a feedstock formulation based on polypropylene for micro-powder soft embossing process of 316L stainless steel micro-channel part

To exploit the potential of micro-system technology, the micro-powder embossing process, a rapid manufacturing production technology of microstructured die mould, was currently being investigated. Present work focused on establishing a suitable binder system for micro-powder embossing process. Multi-component binder systems, comprising of different weight percentages of paraffin wax, stearic acid and polypropylene were investigated. The compatibility between binder constituents was studied by thermogravimetric analysis (TGA) showing a partial miscibility between both components. The feedstock comprised of 316L stainless steel powder, and a wax-based thermoplastic binder was used to achieve a feedstock with 60 to 68 vol.% powder loading. The degradation temperature of binders was determined by using TGA and flow behaviour through rheometer. Homogeneity of the feedstock was verified by using TGA and scanning electron microscopy. Then, hot embossing was done, and it was found that the feedstock having solid loading up to 68 vol.% were successfully embossed, and components were without physical defects. The polymeric part was driven off by thermal debinding using a thermal cycle designed on the basis of a thermogravimetric study of the binder. Finally, the vacuum sintering of the parts allows high quality parts to be obtained.     

Experimental analysis and numerical simulation of sintered micro-fluidic devices using powder hot embossing process

For the past 2 years, interest in manufacturing technologies based on micro-fluidic systems has been continuously increasing. Today, micro-fluidic systems are used in numerous biomedical and pharmaceutical applications. Micro-fluidics cannot be thought about separately without advances in micro- and nano-fabrication. Investigations based on experiments, finite element modelling and simulations of powder hot embossing process (PHE) were performed to optimise the sintering step and processing parameters of micro-fluidic components. The model pertaining to thermo-elasto-viscoplastic behaviour was identified for 316L stainless steel powders. In this regard, different material properties such as sintering stress, bulk, and shearing viscosities were identified by inverse analysis from present dilatometer measurements using beam-bending and free sintering tests. The identification of materials was performed for various powder volume loadings and kinetic rates for different 316L elaborated feedstock, and the parameters were obtained as functions of relative density. The initial inhomogeneity due to the PHE process has been taken into account in the sintering simulation, as it affects the final shrinkage of the sintered components. The solid-state sintering simulations were investigated for various final sintering temperatures and kinetic rates to obtain high and homogeneous relative density distributions, achieve isotropic shrinkage and optimise the sintering process parameters. The numerical simulations were realised based on the identified parameters on a 3D micro-structured specimen with an associated rectangular plate support elaborated by PHE; this allowed a comparison between the numerical predictions and the experimental results for the sintering stage. The finite element simulation results of the sintering stage with a micro-fluidic structured component at a high final temperature (1360 °C) are in excellent agreement with the results of the experiments. The comparison of the simulation and experimental results validated the identified and implemented physical model and proposed methodologies.     

Synthesis of copolymer from 1,3,5‐trioxane and 1,3‐dioxolane catalyzed by Maghnite‐H+

Copolymers (polyoxymethylene) were prepared by cationic copolymerization of 1,3,5‐trioxane (TOX) with 1,3‐dioxolane (DOX) in the presence of Maghnite‐H⁺ (Mag‐H⁺) in solution. Maghnite is a Montmorillonite sheet silicate clay, with exchanged protons to produce Mag‐H⁺. Various techniques, including ¹H‐NMR, ¹³C‐NMR, FT‐IR spectroscopy, and Ubbelohde viscometer were used to elucidate structural characteristics properties of the resulting copolymers. The influence of the amount of catalyst, of dioxolane (DOX), temperature, solvent, and time of copolymerization on yield and on intrinsic viscosity of copolymers was studied. The yield of copolymerization depends on the amount of Mag‐H⁺ used and the reaction time. We also propose mechanisms involved in the synthesis of copolymer (polyoxymethylene). © 2009 Wiley Periodicals, Inc. J Appl Polym Sci, 2010