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.     

An approach to the problem of regional accent in automatic speech recognition

An approach to the problem of inter-speaker variability in automatic speech recognition is described which exploits systematic vowel differences in a two-stage process of adaptation to individual speaker characteristics. In stage one, an accent identification procedure selects one of four gross regional English accents on the basis of vowel quality differences within four calibration sentences. In stage two, an adjustment procedure shifts the regional reference vowel space onto the speaker's vowel space as calculated from the accent identification data. Results for 58 speakers from the four regional accent areas are presented.     

Feature-based modeling for automatic mesh generation

Automatic meshing algorithms for finite element analysis are based on a computer understanding of the geometry of the part to be discretized. Current mesh generators understand the part as either a boundary representation, an octree, or a point set. A higher-level understanding of the part can be achieved by associating engineering significance and engineering data, such as loading and boundary conditions, with generic shapes in the part. This technique, called feature-based modeling, is a popular approach to integrating computer-aided design (CAD) and computer-aided manufacturing through the use of machinable shapes in the CAD model. It would seem that feature-based design also could aid in the finite element mesh generation process by making engineering information explicit in the model.This paper describes an approach to feature-based mesh generation. The feature representation of a fully functioning feature-based system that does automatic process planning and inspection was extended to include finite element mesh generation. This approach is based on a single feature representation that can be used for design, finite element analysis, process planning, and inspection of prismatic parts. The paper describes several advantages that features provide to the meshing process, such as improved point sets and a convenient method of simplifying the geometry of the model. Also discussed are possible extensions to features to enhance the finite element meshing process.     

中文试题模糊自动分类

试题库是在线测试的基础,中文试题的自动分类技术可以提高试题的检索速度和组卷的准确性.中文试题所包含的关键字较少,关键字出现的频度难以掌握,语法结构相似性较大并固定等,利用试题解析算法将试题分解成关键字库和冗字库,建立关键字矩阵,考虑试题的相似性,引入模糊聚类分析技术,对关键字库进行聚类,得到各个主题的类别,通过计算优化后的隶属度对试题进行分类.     

A database for automatic classification of forest species

Forest species can be taxonomically divided into groups, genera, and families. This is very important for an automatic forest species classification system, in order to avoid possible confusion between species belonging to two different groups, genera, or families. A common problem that researchers in this field very often face is the lack of a representative database to perform their experiments. To the best of our knowledge, the experiments reported in the literature consider only small datasets containing few species. To overcome this difficulty, we introduce a new database of forest species in this work, which is composed of 2,240 microscopic images from 112 different species belonging to 2 groups (Hardwoods and Softwoods), 85 genera, and 30 families. To gain better insight into this dataset, we test three different feature sets, along with three different classifiers. Two experiments were performed. In the first, the classifiers were trained to discriminate between Hardwoods and Softwoods, and in the second, they were trained to discriminate among the 112 species. A comprehensive set of experiments shows that the tuple Support Vector Machine (SVM) and Local Binary Pattern (LBP) achieved the best performance in both cases, with a recognition rate of 98.6 and 86.0% for the first and second experiments, respectively. We believe that researchers will find this database a useful tool in their work on forest species recognition. It will also make future benchmarking and evaluation possible. This database will be available for research purposes upon request to the VRI-UFPR.     

Stochastic modeling western paintings for effective classification

As one of the most important cultural heritages, classical western paintings have always played a special role in human live and been applied for many different purposes. While image classification is the subject of a plethora of related publications, relatively little attention has been paid to automatic categorization of western classical paintings which could be a key technique of modern digital library, museums and art galleries. This paper studies automatic classification on large western painting image collection. We propose a novel framework to support automatic classification on large western painting image collections. With this framework, multiple visual features can be integrated effectively to improve the accuracy of identification process significantly. We also evaluate our method and its competitors based on a large image collection. A careful study on the empirical results indicates the approach enjoys great superiority over the state-of-the-art approaches in different aspects.     

Electricity clustering framework for automatic classification of customer loads

Clustering in energy markets is a top topic with high significance on expert and intelligent systems. The main impact of is paper is the proposal of a new clustering framework for the automatic classification of electricity customers’ loads. An automatic selection of the clustering classification algorithm is also highlighted. Finally, new customers can be assigned to a predefined set of clusters in the classification phase. The computation time of the proposed framework is less than that of previous classification techniques, which enables the processing of a complete electric company sample in a matter of minutes on a personal computer. The high accuracy of the predicted classification results verifies the performance of the clustering technique. This classification phase is of significant assistance in interpreting the results, and the simplicity of the clustering phase is sufficient to demonstrate the quality of the complete mining framework.     

Conflicts classification and solving for collaborative feature modeling

Based on the analysis of feature modeling activities in a collaborative environment, a definition and a classification of concurrency conflicts have been presented. A feature adjustment method is proposed to solve the conflicts, an enhanced naming mechanism for the collaborative feature modeling to preserve the design intentions, and a process for the non-locked multi-client collaborative design. The algorithms have been implemented in a prototype system integrating C++, Java3D and VRML, and CORBA technologies. Flexibility and efficiency in collaborative feature modeling environment have been achieved in our system.     

A framework for modeling and evaluating automatic semantic reconciliation

The introduction of the Semantic Web vision and the shift toward machine understandable Web resources has unearthed the importance of automatic semantic reconciliation. Consequently, new tools for automating the process were proposed. In this work we present a formal model of semantic reconciliation and analyze in a systematic manner the properties of the process outcome, primarily the inherent uncertainty of the matching process and how it reflects on the resulting mappings. An important feature of this research is the identification and analysis of factors that impact the effectiveness of algorithms for automatic semantic reconciliation, leading, it is hoped, to the design of better algorithms by reducing the uncertainty of existing algorithms. Against this background we empirically study the aptitude of two algorithms to correctly match concepts. This research is both timely and practical in light of recent attempts to develop and utilize methods for automatic semantic reconciliation.Received: 6 December 2002, Accepted: 15 September 2003, Published online: 19 December 2003Edited by: V. Atluri.     

Experiences on data quality in automatic tissue classification

Automatic intensity-based tissue classification sets requirements for the quality of multispectral magnetic resonance (MR) images. Tests for evaluating the separability of tissue classes, and on the other hand class distances required to obtain reliable classification, are presented in this study. Intraslice, interslice and interpatient training schemes for 5-nn classification were considered. Interslice training was utilized in classification of images from 10 patients with ischemic stroke giving results of satisfactory but highly variable quality. Based on the experience with these data sets, similar tests are recommended before imaging a large patient series in order to avoid extra manual work and to obtain reliable classification results.     

A modular framework for the automatic classification of chromosomes in Q-band images

The manual analysis of the karyogram is a complex and time-consuming operation, as it requires meticulous attention to details and well-trained personnel. Routine Q-band laboratory images show chromosomes that are randomly rotated, blurred or corrupted by overlapping and dye stains. We address here the problem of robust automatic classification, which is still an open issue. The proposed method starts with an improved estimation of the chromosome medial axis, along which an established set of features is then extracted. The following novel polarization stage estimates the chromosome orientation and makes this feature set independent on the reading direction along the axis. Feature rescaling and normalizing techniques take full advantage of the results of the polarization step, reducing the intra-class and increasing the inter-class variances. After a standard neural network based classification, a novel class reassignment algorithm is employed to maximize the probability of correct classification, by exploiting the constrained composition of the human karyotype. An average 94% of correct classification was achieved by the proposed method on 5474 chromosomes, whose images were acquired during laboratory routine and comprise karyotypes belonging to slightly different prometaphase stages. In order to provide the scientific community with a public dataset, all the data we used are publicly available for download.     

ECM-aware cell-graph mining for bone tissue modeling and classification

Pathological examination of a biopsy is the most reliable and widely used technique to diagnose bone cancer. However, it suffers from both inter- and intra- observer subjectivity. Techniques for automated tissue modeling and classification can reduce this subjectivity and increases the accuracy of bone cancer diagnosis. This paper presents a graph theoretical method, called extracellular matrix (ECM)-aware cell-graph mining, that combines the ECM formation with the distribution of cells in hematoxylin and eosin stained histopathological images of bone tissues samples. This method can identify different types of cells that coexist in the same tissue as a result of its functional state. Thus, it models the structure-function relationships more precisely and classifies bone tissue samples accurately for cancer diagnosis. The tissue images are segmented, using the eigenvalues of the Hessian matrix, to compute spatial coordinates of cell nuclei as the nodes of corresponding cell-graph. Upon segmentation a color code is assigned to each node based on the composition of its surrounding ECM. An edge is hypothesized (and established) between a pair of nodes if the corresponding cell membranes are in physical contact and if they share the same color. Hence, multiple colored-cell-graphs coexist in a tissue each modeling a different cell-type organization. Both topological and spectral features of ECM-aware cell-graphs are computed to quantify the structural properties of tissue samples and classify their different functional states as healthy, fractured, or cancerous using support vector machines. Classification accuracy comparison to related work shows that the ECM-aware cell-graph approach yields 90.0% whereas Delaunay triangulation and the simple cell-graph approach achieves 75.0 and 81.1% accuracy, respectively.     

On the automatic classification of app reviews

App stores like Google Play and Apple AppStore have over 3 million apps covering nearly every kind of software and service. Billions of users regularly download, use, and review these apps. Recent studies have shown that reviews written by the users represent a rich source of information for the app vendors and the developers, as they include information about bugs, ideas for new features, or documentation of released features. The majority of the reviews, however, is rather non-informative just praising the app and repeating to the star ratings in words. This paper introduces several probabilistic techniques to classify app reviews into four types: bug reports, feature requests, user experiences, and text ratings. For this, we use review metadata such as the star rating and the tense, as well as, text classification, natural language processing, and sentiment analysis techniques. We conducted a series of experiments to compare the accuracy of the techniques and compared them with simple string matching. We found that metadata alone results in a poor classification accuracy. When combined with simple text classification and natural language preprocessing of the text—particularly with bigrams and lemmatization—the classification precision for all review types got up to 88–92 % and the recall up to 90–99 %. Multiple binary classifiers outperformed single multiclass classifiers. Our results inspired the design of a review analytics tool, which should help app vendors and developers deal with the large amount of reviews, filter critical reviews, and assign them to the appropriate stakeholders. We describe the tool main features and summarize nine interviews with practitioners on how review analytics tools including ours could be used in practice.     

Advances in classification for land cover mapping using SPOT HRV imagery

Abstract

High-resolution data from the HRV (High Resolution Visible) sensors onboard the SPOT-1 satellite have been utilized for mapping semi-natural and agricultural land cover using automated digital image classification algorithms. Two methods for improving classification performance are discussed. The first technique involves the use of digital terrain information to reduce the effects of topography on spectral information while the second technique involves the classification of land-cover types using training data derived from spectral feature space. Test areas in Snowdonia and the Somerset Levels were used to evaluate the methodology and promising results were achieved. However, the low classification accuracies obtained suggest that spectral classification alone is not a suitable tool to use in the mapping of semi-natural cover types.     

Exploiting articulatory features for pitch accent detection

Articulatory features describe how articulators are involved in making sounds. Speakers often use a more exaggerated way to pronounce accented phonemes, so articulatory features can be helpful in pitch accent detection. Instead of using the actual articulatory features obtained by direct measurement of articulators, we use the posterior probabilities produced by multi-layer perceptrons （MLPs） as articulatory features. The inputs of MLPs are frame-level acoustic features pre-processed using the split temporal context-2 （STC-2） approach. The outputs are the posterior probabilities of a set of articulatory attributes. These posterior probabilities are averaged piecewise within the range of syllables and eventually act as syllable-level articulatory features. This work is the first to introduce articulatory features into pitch accent detection. Using the articulatory features extracted in this way, together with other traditional acoustic features, can improve the accuracy of pitch accent detection by about 2%.     

Document retrieval by means of an automatic classification algorithm for citations

The applicability of an automatic classification technique to information retrieval was investigated. A modified version of the Schiminovich algorithm was used to classify articles in the data base, utilizing citations found in their bibliographies and a “triggering file” of bibliographically related papers. Classes of articles were formed by comparing the citations in the articles with members of the triggering file. Those papers with similar citing patterns formed a group; citations occurring sufficiently often within the papers of a group formed a bibliography. Bibliographies in turn became new triggering files in an iterative procedure.Results of this nontraditional method were compared with those of retrieval by standard American Institute of Physics (AIP) subject analysis of the same material. A combination of the two methods was also investigated to test the hypothesis that one could be used to augment or refine the other.Nine cooperating physicists defined queries in terms of (1) the AIP classification scheme and (2) a list of articles likely to be cited in current relevant literature. The data base was 18 months of 1971–1972 physics journal articles (AIP SPIN tapes). The chief means of evaluating the results of the three retrieval approaches was a comparison of recall and precision values obtained from relevance judgments by the physicists.Average precision for the AIP subject analysis was 17 per cent and for the citation processing 62 per cent. Based on an assumption of 100 per cent recall for the AIP analysis, average recall for the citation processing was 45 per cent. All but one physicist would prefer to have both the citation and subject approaches available for information retrieval. Sometimes, only a few relevant articles are desired; at other times comprehensiveness is necessary. However, if only one method were available, all except one participant would choose the subject approach. They felt that one must be able to retrieve all the relevant articles, even if that would mean examining 25–30 irrelevant notices for every relevant one. Several expressed the notion that looking at a file with a very high percentage of relevant articles made one wonder what was missing.     

Methodology for automatic bioacoustic classification of anurans based on feature fusion

The automatic recognition of anurans by their calls provides indicators of ecosystem health and habitat quality. This paper presents a new methodology for the acoustic classification of anurans using a fusion of frequency domain features, Mel and Linear Frequency Cepstral Coefficients (MFCCs and LFCCs), with time domain features like entropy and syllable duration through intelligent systems. This methodology has been validated in three databases with a significant number of different species proving the strength of this approach. First, the audio recordings are automatically segmented into syllables which represent different anuran calls. For each syllable, both types of features are computed and evaluated separately as in previous works. In the experiments, a novel data fusion method has been used showing an increase of the classification accuracy which achieves an average of 98.80% ± 2.43 in 41 anuran species from AmphibiaWeb database, 96.90% ± 3.57 in 58 frogs from Cuba and 95.48% ± 4.97 in 100 anurans from southern Brazil and Uruguay; reaching a classification rate of 95.38% ± 5.05 for the aggregate dataset of 199 species.