双模型语音识别中的听视觉合成和模型同步异步性实验研究

研究了双模型语音识别系统中前合成和后合成两种听觉视觉合成方法；同时在后合成方法中引入了考虑听觉和视觉同步异步特点的复合模型。仿真实验证明了在声学噪音环境下，后合成方法能够带来比较理想的识别效果；考虑听觉和视觉同步异步性的模型可以有效地提高识别率。     

Identification Facilitation for Emotionally Arousing Verbs During the Attentional Blink.

The present study aimed to examine affective modulation of the "attentional blink" effect during rapid serial visual presentation (RSVP). Pleasant, neutral, and unpleasant written verbs were used as a 2nd target (T2) in an 8.6-Hz RSVP paradigm. Pronounced effects of 1st target (T1)-T2 stimulus onset asynchrony (SOA) were found, showing reduced report accuracy for 232- and 464-ms SOAs. Affectively arousing (pleasant and unpleasant) T2s were associated with enhanced accuracy compared with neutral T2s specifically during short (232 ms) SOAs. In contrast, pleasant and unpleasant T2s rated low in terms of emotional arousal did not show this enhancement. These results suggest that affectively arousing information is selected preferentially from a temporal stream, facilitating processes such as working memory consolidation and action. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

基于滑动聚类的多传感器异步信息融合方法

传感器是智能检测和自动化装置中重要的部件组成,为了解决多传感器异步数据下的融合难题,提出了一种创新的基于滑动聚类的多传感器异步信息融合方法。首先引入了K-means聚类方法去容忍异步问题,主要利用曲线拟合给出一种简易的快速的判定法则以便实现实时聚类方法中的k值计算;其次设计了聚类滤波核从而在时空域上形成融合滑动管道,让数据的变化一直维持在一个能接受的误差之内,完整实现了实时多传感器信息融合方法。最后实验验证了设计的聚类融合方法的正确性以及合理性,实验证明了SC-MSIF方法是正确可行的,且在实时性方面具有较好的表现,相比较EKF和MEAN方法,SC-MSIF方法的RMSE误差减少了47.8%、36.3%,同时无人机中多传感器融合实际测试结果也较为优异。     

The time course of competition for attention: Attention is initially labile.

Competition for attention between 2 written words was investigated by presenting the words briefly in a single stream of distractors (Experiment 1) or in different streams (Experiment 2- 6), using rapid serial visual presentation at 53 ms/item. Stimulus onset asynchrony (SOA) was varied from 0 to 213 ms. At all SOAs there was strong competition, but which word was more likely to be reported shifted markedly with SOA. At SOAs in the range of 13-53 ms the second word was more likely to be reported, but at 213 ms, the advantage switched to the first word, as in the attentional blink. A 2-stage competition model of attention is proposed in which attention to a detected target is labile in Stage 1. Stage 1 ends when one target is identified, initiating a serial Stage 2 process of consolidation of that target. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

The relationship between attentional blink and psychological refractory period.

This study examined the relationship between attentional blink (AB) and psychological refractory period (PRP) using a conventional AB procedure combined with a requirement of speeded responses to the second target (T2). Experiments 1 and 2 showed that, as with PRP, memory retrieval of targets is not a necessary condition for the occurrence of AB in terms of accuracy and that AB occurred in the speed data. Experiment 3 further indicated that the PRP-like speed data were not due to the first target serving as a warning signal that triggered preparation of responses to T2. Experiment 4 manipulated T2 stimulus intensity to be normal or low. Results showed an underadditive interaction between stimulus intensity and lag position in the speed data, whereas an overadditive interaction was found in the accuracy data, suggesting 2 sources of interference leading to AB: the single-channel bottleneck and resource-limited visual-processing capacity. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

循环爆破作用下岩体的不同步累积损伤测试方法

为了提高循环爆破载荷作用下岩体累积损伤测试方法的精确度、预测结果的可信值以及控制标准的规范化,提出不同步累积损伤测试方法。采用理论分析与现场实践相结合的方法,基于5次循环爆破载荷作用下不同测孔深度处声波速度的变化规律,依据损伤度与声波速度的关系式,对比分析2种测试方法的累积损伤效应。研究结果表明:循环爆破载荷仅对测孔孔口至1.6 m处岩体区域产生损伤效应,并且垂直于掌子面的剖面区域所受的累积损伤最为严重;不同步测试方法所得的岩体累积损伤效应更加明显,且累积损伤-爆破次数关系曲线更加光滑,由此证明,不同步累积损伤测试方法能够提高测试精度,增加拟合曲线的可信程度,不仅为支护厚度提供依据,而且为支护方向提供参考。     

三峡水库汛期水沙异步特性及沙峰输移时间研究

以三峡水库运用以来2003—2018年库区主要测站的水沙数据为基础,对上游来沙入库时刻水沙异步特性进行了统计分析,讨论了不同水位级情况下典型场次洪水过程中沙峰输移时间与不同影响因素之间的关系,并利用回归分析法得到了拟合程度较好的沙峰输移时间回归公式。结果表明:洪峰流量>30 000 m³/s的40场典型洪水过程,入库时刻沙峰超前、同步、滞后洪峰的场次比例分别为42.5%、42.5%、15%;金沙江下游向家坝等梯级水库蓄水运用后,沙峰在三峡库区的传播时间明显增长。由于水库不同运用时期汛期坝前水位变化范围不同,采用了分水位级的方式对库区沙峰输移时间与不同影响因素的相关性进行讨论,进而使用逐步回归的方法分别建立了拟合效果较好的沙峰传播时间公式,可为水库沙峰调度提供技术支撑。     

模拟阅读脑-机接口异步化研究

“模拟阅读”脑机接口(BCI)工作在同步模式,而实际中使用者希望能在“工作/非工作”状态间自由切换,即异步化,针对该问题提出了利用闭眼固定时间的脑电信号作为两种状态间转换开关的方法。首先设计了实验方案;然后对采集的脑电图(EEG)信号分别在时域和频域进行特征提取,对时域特征利用支持向量机(SVM)和K-means分类器进行分类,对频域特征用SVM分类。时域最高识别率分别为91.25%和89.17%,平均分类所需时间分别为1.89s和0.11s,频域最高识别率和平均识别率分别为86.25%和81.875%。实验结果表明该实验模式能实现两种状态自由切换的目的。     

With great expectations, can two "wrongs" prime a "right"?

The proportion of related prime-target pairs (relatedness proportion, RP) and prime-target stimulus onset asynchrony (SOA) was varied to determine the involvement of strategic priming mechanisms in the reduction in semantic priming that occurs when a target follows an unmasked prime that itself receives immediate repetition priming from a masked prime. At 300-ms and 1,200-ms SOAs, (a) strategic semantic priming was operating, in that priming from a nonrepeated prime increased as RP increased from .25 to .75, and (b) for both RPs, prime repetition reduced semantic priming. At a 167-ms SOA, (a) priming from a nonrepeated prime was unaffected by RP, suggesting that strategic priming was not operating, and (b) for both RPs, prime repetition did not reduce semantic priming. Because prime repetition did not reduce priming at the 167-ms SOA (when only spreading activation should have been mediating semantic priming), the reduction in semantic priming produced by prime repetition is not evidence against spreading activation automaticity. Possible mechanisms through which prime repetition reduces semantic priming are discussed. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

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%.