汉语方言的集成决策辨识

针对方言辨识系统分类决策能力较弱的问题,提出一种基于集成学习的方言辨识方法。该方法将高斯模型与语言模型组成的系统作为一组基分类器,然后根据这组基分类器所得分类结果的加权组合来决定方言语音所属类别。实验结果表明,新的集成决策分类方法不仅可以大大提高系统的辨识精度,而且有效地解次训练样本数目和模型参数之间的矛盾。     

Understanding U.S. regional linguistic variation with Twitter data analysis

We analyze a Big Data set of geo-tagged tweets for a year (Oct. 2013–Oct. 2014) to understand the regional linguistic variation in the U.S. Prior work on regional linguistic variations usually took a long time to collect data and focused on either rural or urban areas. Geo-tagged Twitter data offers an unprecedented database with rich linguistic representation of fine spatiotemporal resolution and continuity. From the one-year Twitter corpus, we extract lexical characteristics for twitter users by summarizing the frequencies of a set of lexical alternations that each user has used. We spatially aggregate and smooth each lexical characteristic to derive county-based linguistic variables, from which orthogonal dimensions are extracted using the principal component analysis (PCA). Finally a regionalization method is used to discover hierarchical dialect regions using the PCA components. The regionalization results reveal interesting linguistic regional variations in the U.S. The discovered regions not only confirm past research findings in the literature but also provide new insights and a more detailed understanding of very recent linguistic patterns in the U.S.     

To What Extent are Surnames Words? Comparing Geographic Patterns of Surname and Dialect Variation in the Netherlands

Since the early studies by Sokal (1988) and Cavalli-Sforza etal. (1989), there has been an increasing interest in depictingthe history of human migrations by comparing genetic and linguisticdifferences that mirror different aspects of human history.Most of the literature concerns continental or macroregionalpatterns of variation, while regional and microregional scaleswere investigated less successfully. In this article we concentrateon the Netherlands, an area of only 40,000 km². The focus of the article is on the analysis of surnames, whichhave been proven to be reliable genetic markers since in patrilinealsystems they are transmitted—virtually unchanged—alonggenerations, similar to a genetic locus on the Y-chromosome.We shall compare their distribution to that of dialect pronunciations,which are clearly culturally transmitted (children learn oneof the linguistic varieties they are exposed to, normally thatof their peers in the same area or that of their families).Since surnames, at the time of their introduction, were wordssubject to the same linguistic processes that otherwise resultin dialect differences, one might expect the distribution ofsurnames to be correlated with dialect pronunciation differences.But we shall argue that once the collinear effects of geographyon both genetics and cultural transmission are taken into account,there is in fact no statistically significant association betweenthe two. We show that surnames cannot be taken as a proxy fordialect variation, even though they can be safely used as aproxy to Y-chromosome genetic variation. We work primarily with regression analyses, which show thatboth surname and dialect variation are strongly correlated withgeographic distance. In view of this strong correlation, wefocus on the residuals of the regression, which seeks to explaingenetic and linguistic variation on the basis of geography (wheregeographic distance is the independent variable, and surnamediversity or linguistic diversity is the dependent variable).We then seek a more detailed portrait of the geographic patternsof variation by identifying the ‘barriers’ (namelythe areas where the residuals are greatest) by applying theMonmonier algorithm. We find the results historically and geographically insightful,hopefully leading to a deeper understanding of the role of thelocal migrations and cultural diffusion that are responsiblefor surname and dialect diversity.     

基于中文拼音输入法数据的汉语方言词汇自动识别

方言研究领域中的语音研究、词汇研究及语法研究是方言研究的三个重要组成部分,如何识别方言词汇,是方言词汇研究首要的环节。目前,汉语方言词汇研究的语料收集与整理主要通过专家人工整理的形式进行,耗时耗力。 随着信息技术的发展,人们的交流广泛通过网络进行,而输入法数据包含海量的语料资源以及地域信息,可以帮助进行方言词汇语料的自动发现。然而,目前尚没有文献研究如何利用拼音输入法数据对方言词汇进行系统化分析,因此在本文中,我们探讨借助中文输入法的用户行为来自动发现各地域方言词汇的方法。特别的,我们归纳得到输入法数据中表征方言词汇的两类特征,并基于对特征的不同组合识别方言词汇。最后我们通过实验评价了两类特征的不同组合方法对方言词汇识别效果的影响。     

贵阳方言工厂指令识别

针对贵阳工厂环境下口头任务对接缺乏依据性、出现事故难于追责的问题,引入深度学习模型改善贵阳方言工厂指令识别效果.自制贵阳方言工厂指令数据集,搭建指令识别系统,依次训练六种模型,其中包括拥有9层隐藏层的深度神经网络.在同一测试集下,系统随训练的进行逐渐提升性能,在DNN模型下识别错误率降至最低,远低于单音素模型识别错误率...     

DTZH1505：大规模开源中文普通话语音库

近年来,深度学习在语音识别领域取得了突破性进展,并推动语音识别技术广泛应用到人们的日常生活中。语音识别模型的进一步优化需要更大规模标定数据的驱动,然而,目前开源的语音数据集规模仍太小,语料多为偏向书面用语的新闻类长文本。针对人机交互、智能客服等热门语音识别应用,通过众包模式采集朗读式语音,构建并开源了迄今为止最大规模的中文普通话语音数据集DTZH1505。数据集记录了6?408位来自中国八大方言地域、33个省份的说话人的自然语音,时长达1?505?h,语料内容涵盖社交聊天、人机交互、智能客服以及车载命令等,可广泛用于语料库语言学、会话分析、语音识别、说话人识别等研究。开展一系列基准语音识别实验,实验结果表明：相较于同规模中文语音数据集aishell2,基于此数据集训练的语音识别模型效果更好。     

Sichuan dialect speech recognition with deep LSTM network

In speech recognition research,because of the variety of languages,corresponding speech recognition systems need to be constructed for different languages.Especially in a dialect speech recognition system,there are many special words and oral language features.In addition,dialect speech data is very scarce.Therefore,constructing a dialect speech recognition system is difficult.This paper constructs a speech recognition system for Sichuan dialect by combining a hidden Markov model(HMM)and a deep long short-term memory(LSTM)network.Using the HMM-LSTM architecture,we created a Sichuan dialect dataset and implemented a speech recognition system for this dataset.Compared with the deep neural network(DNN),the LSTM network can overcome the problem that the DNN only captures the context of a fixed number of information items.Moreover,to identify polyphone and special pronunciation vocabularies in Sichuan dialect accurately,we collect all the characters in the dataset and their common phoneme sequences to form a lexicon.Finally,this system yields a 11.34%character error rate on the Sichuan dialect evaluation dataset.As far as we know,it is the best performance for this corpus at present.     

“方言同音字汇”自动生成软件①的设计及实现

“方言同音字汇”整理是方言调查的基础性工作,靠手工制作十分繁难。该文论述了“方言同音字汇”自动生成软件的设计原理及实现过程。软件的主要功能是,根据用户事先给定的韵、声、调排序依据和排序顺序,对已经录入的方言字表进行排序,排序技术采用对应韵、声、调与字表所有字目的一个四重循环,最终生成“同音字汇竖排表”。此外,该文对软件的实用性能进行了分析,并对软件的应用进行了一定的说明。实践证明,该软件完全能够满足方言调查实用化的需求。     

藏语机读音标SAMPA_ST的设计

该文选取具有代表意义的藏语卫藏方言的拉萨话、安多方言的夏河话以及康方言的德格话进行语言调查;整理归纳藏语三大方言音系,包括单辅音、复辅音、单元音、复合元音和辅音韵尾,以及三大方言声调;依照SAMPA的规则建立适合于藏语三大方言的机读音标,并设计了SAMPA_ST的自动标注系统,实现文音转换功能,为语音的韵律特征分析和语音工程的研究提供依据。     

Analysis and modeling of acoustic information for automatic dialect classification

A primary challenge in the field of automatic speech recognition is to understand and create acoustic models to represent individual differences in their spoken language. Individual’s age, gender; their speaking styles influenced by their dialect may be few of the reasons for these differences. This work investigates the dialectal differences by measuring the analysis of variance of acoustic features such as, formant frequencies, pitch, pitch slope, duration and intensity for vowel sounds. This paper attempts to discuss methods to capture dialect specific knowledge through vocal tract and prosody information extracted from speech that can be utilized for automatic identification of dialects. Kernel based support vector machine is utilized for measuring the dialect discriminating ability of acoustic features. For the spectral feature shifted delta cepstral coefficients along with Mel frequency cepstral coefficients gives a recognition performance of 66.97 %. Combination of prosodic features performs better with a classification score of 74 %. The model is further evaluated for the combination of spectral and prosodic feature set and achieves a classification accuracy of 88.77 %. The proposed model is compared with the human perception of dialects. The overall work is based on four dialects of Hindi; one of the world’s major languages.     

North American English Vowels: A Factor-analytic Perspective

Previous studies of American English have identified a numberof robust patterns involving the vowel system, such as the NorthernCities Chain Shift and the Southern Vowel Shift. These studiesprimarily employ methods which treat separately the phoneticproperties of specific vowels as produced by individual speakerswhich are later assembled into complete vowel systems. Whilethis provides a useful picture of production, it is not adequatefor comparison with dialect perception studies, where interpretationof the results often requires some understanding of the correlationsamong linguistic features and between those features and individualtalkers. We conducted a factor analysis of the duration andfirst and second formant frequencies of each of the fourteenvowels produced by forty-eight speakers representing six regionalvarieties of American English and both genders. The data weresubmitted to factor analysis using maximum likelihood estimationand Varimax rotation. Results confirmed significant correlationsbetween regional dialect and acoustic–phonetic propertiesof the vowel systems, although these patterns are complicatedby interactions with gender. These results illustrate the utilityof factor analytic methods in examining systematic variationacross an entire linguistic system such as the vowels.     

The use of the computer in plotting the geographical distribution of dialect items

Concluding comments To date we have produced maps with all responses to each individual question, frequency counts for all primary responses, and individual maps for separate responses that occurred at least 10 times. These individual-response maps were used in drawing the isoglosses on various maps, as illustrated in Maps I–IV. As of January 1976, all the data have been coded, keypunched, proofread, corrected, and stored on magnetic tape. What remains to be done is continuing modification of the programming to produce (1) maps that use variables other than geographical location, (2) statistical analyses of the basic data in a sociolinguistic perspective, (3) further labeling of data to reflect the dialect divisions found in the preliminary analysis, and finally (4) further statistical analyses of data within and along the edges of these dialect subdivisions. Because the initial goal, the identification of clearly regional items, is a somewhat limited, though a necessary preliminary of the entire investigation, we have not yet programmed the subroutines to calculate densities of items in given areas or to compute statistical correlations among variables. Even though it is still necessary, with our present program and data files, to find distributional patterns by eye and draw isoglosses by hand, ultimately the entire operation enables us to accomplish much more in a shorter period of time than would ever have been possible without the computer. Financial support for the research and programming was provided by the Research Council of the Graduate School, University of Missouri-Columbia.     

基于差分特征和高斯混合模型的湖南方言识别

语音的韵律是区分汉语方言的重要语音声学特征,而语音的差分特征是语音韵律的重要体现。采用差分特征ΔMFCC和ΔΔMFCC作为特征参数,用高斯混合模型（GMM）作为训练模型,通过计算测试样本的似然概率来识别方言的类型。用该方法对长沙方言、邵阳方言、衡阳方言和普通话进行了识别研究,并与采用MFCC作为特征参数的识别效果进行了比较。实验结果表明差分特征具有识别率高、抗噪声性能更好等优点。     

The Design of OpenMP Tasks

OpenMP has been very successful in exploiting structured parallelism in applications. With increasing application complexity, there is a growing need for addressing irregular parallelism in the presence of complicated control structures. This is evident in various efforts by the industry and research communities to provide a solution to this challenging problem. One of the primary goals of OpenMP 3.0 is to define a standard dialect to express and efficiently exploit unstructured parallelism. This paper presents the design of the OpenMP tasking model by members of the OpenMP 3.0 tasking sub-committee which was formed for this purpose. The paper summarizes the efforts of the sub-committee (spanning over two years) in designing, evaluating and seamlessly integrating the tasking model into the OpenMP specification. In this paper, we present the design goals and key features of the tasking model, including a rich set of examples and an in-depth discussion of the rationale behind various design choices. We compare a prototype implementation of the tasking model with existing models, and evaluate it on a wide range of applications. The comparison shows that the OpenMP tasking model provides expressiveness, flexibility, and huge potential for performance and scalability.     

面向AI的语音数据采集服务平台的设计与实现

为解决语音AI 的方言语音数据采集存在的数据量不够多、样本分布不均衡等问题,以语音数据收集、标注、数据交叉校验、数据集打包分享为目标,设计开发了一个语音数据采集与服务平台,提供语音数据采集、任务定制、语音与文本数据管理、数据标注、数据检索、数据下载等功能,通过微信小程序和手机APP吸引用户参与有趣的语音游戏,从而实现可定制的语音数据采集、标注、交叉校验等工作,在提升语音数据量的同时,有效解决数据采集过程中的样本分布不均衡问题,提升语音数据在方言人群和地域方面覆盖范围,提升数据质量,助力方言语音识别。     

Type-2 fuzzy hidden Markov models and their application to speech recognition

This paper presents an extension of hidden Markov models (HMMs) based on the type-2 (T2) fuzzy set (FS) referred to as type-2 fuzzy HMMs (T2 FHMMs). Membership functions (MFs) of T2 FSs are three-dimensional, and this new third dimension offers additional degrees of freedom to evaluate the HMMs fuzziness. Therefore, T2 FHMMs are able to handle both random and fuzzy uncertainties existing universally in the sequential data. We derive the T2 fuzzy forward-backward algorithm and Viterbi algorithm using T2 FS operations. In order to investigate the effectiveness of T2 FHMMs, we apply them to phoneme classification and recognition on the TIMIT speech database. Experimental results show that T2 FHMMs can effectively handle noise and dialect uncertainties in speech signals besides a better classification performance than the classical HMMs.     

Optimal Use of Mixed Task and Data Parallelism for Pipelined Computations

This paper addresses optimal mapping of parallel programs composed of a chain of data parallel tasks onto the processors of a parallel system. The input to the programs is a stream of data sets, each of which is processed in order by the chain of tasks. This computation structure, also referred to as a data parallel pipeline, is common in several application domains, including digital signal processing, image processing, and computer vision. The parameters of the performance for such stream processing are latency (the time to process an individual data set) and throughput (the aggregate rate at which data sets are processed). These two criteria are distinct since multiple data sets can be pipelined or processed in parallel. The central contribution of this research is a new algorithm to determine a processor mapping for a chain of tasks that optimizes latency in the presence of a throughput constraint. We also discuss how this algorithm can be applied to solve the converse problem of optimizing throughput with a latency constraint. The problem formulation uses a general and realistic model of intertask communication and addresses the entire problem of mapping, which includes clustering tasks into modules, assigning of processors to modules, and possible replicating of modules. The main algorithms are based on dynamic programming and their execution time complexity is polynomial in the number of processors and tasks. The entire framework is implemented as an automatic mapping tool in the Fx parallelizing compiler for a dialect of High Performance Fortran.     

Gradual representation of shadowed set for clustering gene expression data

Micro array has been a widely used microscopic measurement that accumulates the expression levels of a large number of genes varying over different time points. Cluster analysis more over the concept of bi-clustering provides insight into meaningful information from the correlation of a subset of genes with a subset of conditions. This eventually helps in discovering biologically meaningful clusters over analyzing missing values, imprecision and noise present in micro array data set. Although the concept of fuzzy set is enough to deal with the overlapping nature of the bi-clusters but the use of shadowed set helps in identifying and analyzing the nature of the genes lying in the confusion area of the clusters. In this article, we have suggested a bi-clustering model of the shadowed set with gradual representation of cardinality and named it as Gradual shadowed set for gene expression (GSS-GE) clustering. It identifies the bi-clusters in the core and in the shadowed region and evaluates their biological significance. The excellence of the proposed GSS-GE has been demonstrated by considering three real data sets, namely yeast data, serum data and mouse data set. The performance is compared with Ching Church’s algorithm (CC), Bimax, order preserving sub matrix (OPSM), Large Average Sub matrices (LAS), statistical plaid model and a modified fuzzy co-clustering (MFCC) algorithm. For the mouse data set there is no cluster level analysis of the micro array has been done so far. We have also provided the statistical and biological significance to prove the superiority of the proposed GSS-GE.     

Multi-view fuzzy clustering with minimax optimization for effective clustering of data from multiple sources

Multi-view data clustering refers to categorizing a data set by making good use of related information from multiple representations of the data. It becomes important nowadays because more and more data can be collected in a variety of ways, in different settings and from different sources, so each data set can be represented by different sets of features to form different views of it. Many approaches have been proposed to improve clustering performance by exploring and integrating heterogeneous information underlying different views. In this paper, we propose a new multi-view fuzzy clustering approach called MinimaxFCM by using minimax optimization based on well-known Fuzzy c means. In MinimaxFCM the consensus clustering results are generated based on minimax optimization in which the maximum disagreements of different weighted views are minimized. Moreover, the weight of each view can be learned automatically in the clustering process. In addition, there is only one parameter to be set besides the fuzzifier. The detailed problem formulation, updating rules derivation, and the in-depth analysis of the proposed MinimaxFCM are provided here. Experimental studies on nine multi-view data sets including real world image and document data sets have been conducted. We observed that MinimaxFCM outperforms related multi-view clustering approaches in terms of clustering accuracy, demonstrating the great potential of MinimaxFCM for multi-view data analysis.     

A UML profile for the conceptual modelling of structurally complex data: Easing human effort in the KDD process

ContextDomains where data have a complex structure requiring new approaches for knowledge discovery from data are on the increase. In such domains, the information related to each object under analysis may be composed of a very broad set of interrelated data instead of being represented by a simple attribute table. This further complicates their analysis.ObjectiveIt is becoming more and more necessary to model data before analysis in order to assure that they are properly understood, stored and later processed. On this ground, we have proposed a UML extension that is able to represent any set of structurally complex hierarchically ordered data. Conceptually modelled data are human comprehensible and constitute the starting point for automating other data analysis tasks, such as comparing items or generating reference models.MethodThe proposed notation has been applied to structurally complex data from the stabilometry field. Stabilometry is a medical discipline concerned with human balance. We have organized the model data through an implementation based on XML syntax.ResultsWe have applied data mining techniques to the resulting structured data for knowledge discovery. The sound results of modelling a domain with such complex and wide-ranging data confirm the utility of the approach.ConclusionThe conceptual modelling and the analysis of non-conventional data are important challenges. We have proposed a UML profile that has been tested on data from a medical domain, obtaining very satisfactory results. The notation is useful for understanding domain data and automating knowledge discovery tasks.