An Approach Using Combination of Multiple Features through Sigmoid Function for Speech-Presence/Absence Discrimination

Discriminative training aims at constructing a classifier that is of small scale but has high classification power. One type, Minimum Classification Error (MCE) training, has been used widely in pattern recognition, especially in the speech recognition field. In parallel with this, Relevance Vector Machine (RVM) has attracted many researchers' interest, based on its potential for alleviating the scalability problem of Support Vector Machine (SVM). It has been reported that RVM achieves high classification accuracy with a limited amount of classifier parameters, i.e., relevance vectors. Comparison studies between MCE training and SVM have been done, but not so much between MCE training and RVM. Motivated by this, we conduct theoretical and experimental comparisons of MCE training and RVM. Results show that MCE training is better suited to the development of small-scale but highly discriminative classifiers than its counterpart.

In this paper, we present the theoretical framework of minimum classification error (MCE) training of generalized linear classifiers for text classification. We show that many important text classifiers, either probabilistic or non-probabilistic, can be unified under this framework, and the proposed MCE classifier training approach can be applied to improve the classifier performance. In addition, we describe an effective MCE classifier training algorithm that uses AdaBoost to generate alternative initial classifiers, as opposed to combining multiple classifiers as it is typically used. This method is applied to MCE classifier training to overcome local minimums in optimal classifier parameter search, utilizing the fact that the family of generalized linear classifiers is closed under AdaBoost. Moreover, we extend the loss function in MCE training to incorporate training sample prior distributions to compensate the imbalanced training data distribution in each category. Experimental studies are performed on the text classification tasks, and the significant classification error reductions of 25% - 55% are observed.

This paper considers both the garbage modelling and the discriminative training of a speech recognition system consisting of a multilayer perceptron (MLP) network and hidden Markov models (HMMs). The string level training of the system is based on Minimum Classification Error (MCE) algorithm using especially the ideas recently proposed by Reichl et al. In order to improve the MCE training scheme, they defined a derivative for MCE cost function putting heavier weight to the misclassified utterances than the traditional translated sigmoidal. In this paper, we give a definition of a new cost function, whose derivative has similar properties as the one proposed by Reichl et al. The new and the sigmoidal cost function are compared with the string level (MCE) algorithm. The string recognition rates show that both methods achieve equal performances, but the convergence of the MCE algorithm is a bit faster with the new cost function. Moreover, the performance of two garbage models, the nth best and an interpolative garbage model, are also compared in this paper. The latter one is an approximation of the former one requiring less operations, but achieving comparable performance. The recognition system achieved 93.32% accuracy for test set at best. The test set contained 29188 Finnish digit strings from two environments.

In our previous work, we have proposed a speaker modeling technique using spectral and pitch features for text-independent speaker identification based on Multi-Space Probability Distribution Gaussian Mixture Models (MSD-GMMs). We have presented a maximum likelihood (ML) estimation procedure for the MSD-GMM parameters and demonstrated its high recognition performance. In this paper, we describe an minimum classification error (MCE) training procedure for the MSDGMM speaker models. MCE training is also applied to automatically estimate mixture-dependent stream weights for spectral and pitch streams. The MCE-based MSD-GMM speaker models are evaluated for a text-independent speaker identification task. Experimental results show that MCE training of the MSD-GMM parameters significantly reduces identification errors and system performance is further improved by appropriately weighting spectral and pitch streams using MCE training.

The minimum classification error (MCE) framework for discriminative training is a simple and general formalism for directly optimizing recognition accuracy in pattern recognition problems. The framework applies directly to the optimization of hidden Markov models (HMMs) used for speech recognition problems. However, few if any studies have reported results for the application of MCE training to large-vocabulary, continuous-speech recognition tasks. This article reports significant gains in recognition performance and model compactness as a result of discriminative training based on MCE training applied to HMMs, in the context of three challenging large-vocabulary (up to 100 k word) speech recognition tasks: the Corpus of Spontaneous Japanese lecture speech transcription task, a telephone-based name recognition task, and the MIT Jupiter telephone-based conversational weather information task. On these tasks, starting from maximum likelihood (ML) baselines, MCE training yielded relative reductions in word error ranging from 7% to 20%. Furthermore, this paper evaluates the use of different methods for optimizing the MCE criterion function, as well as the use of precomputed recognition lattices to speed up training. An overview of the MCE framework is given, with an emphasis on practical implementation issues

Objective To discuss the predictive value of multiple ultrasonic features combination in diagnosis of thyroid papillary carcinoma (PTC) more than 1.0 cm in diameter. Methods The ultrasonic features of 258 PTC nodules from 251 patients and 207 nodular goiter (NG) nodules from 190 patients in the First People′s Hospital of Hangzhou were retrospectively analyzed. All the nodules were confirmed by pathological examination after surgery. The ultrasonic features included the shape of nodules, internal echo, anteroposterior/transverse diameter ratio (A/T), and microcalcification. The χ2 test was used to analyze the differences of ultrasonic features between PTC and NG. Multi-variate analyses (Logistic regression) was used to analyze the predictive risk ultrasonic features of PTC. The sensitivity and specificity of ultrasonic features were analyzed based on the gold standard of pathological results. Results There were significantly differences between 258 PTC nodules and 207 NG nodules in irregular shape, hypoechogenicity, A/T>1 and microcalcification (χ2 values were 121.511, 105.411, 41.483, 121.072, all P 1 and microcalcification were risk ultrasonic features of PTC. And their OR values were 5.013 (95%CI 2.919-8.610), 5.811 (95%CI 3.411-9.901), 15.399 (95%CI 7.576-31.301), 4.141 (95%CI 1.687-10.164) respectively. The sensitivity and specificity of single ultrasonic feature were 26.0%-79.5% and 71.5%-96.1%; the sensitivity and specificity of two ultrasonographic features combination were 11.2%-57.0% and 92.3%-99.0%; and the sensitivity and specificity of three or four ultrasonographic features combination were 8.1%-31.8% and 99.0%-99.5%. Conclusions Irregular shape, hypoechogenicity, A/T>1 and microcalcification of thyroid neoplasm are important ultrasonic features of PTC. Although the sensitivity of single ultrasonic feature in diagnosing PTC is higher than that of multiple features combination, it has a lower specificity. Therefore, combination of multiple ultrasonographic features can improve the specificity in diagnosing PTC and reduce the misdiagnosis of PTC. Key words: Ultrasonography; Thyroid neoplasm; Nodular goiter

Keyword spotting can be formulated as a non-uniform error automatic speech recognition (ASR) problem. It has been demonstrated [1] that this new formulation with the nonuniform MCE training technique can lead to improved system performance in keyword spotting applications. In this paper, we demonstrate that deep neural networks (DNNs) can be successfully trained on the non-uniform minimum classification error (MCE) criterion which weighs the errors on keywords much more significantly than those on non-keywords in an ASR task. The integration with a DNN-HMM system enables modeling of multi-frame distributions, which conventional systems find difficult to accomplish. To further improve the performance, more confusable data is generated by boosting the likelihood of the sentences that have more errors. The keyword spotting system is implemented within a weighted finite state transducer (WFST) framework and the DNN is optimized using standard backpropagation and stochastic gradient decent. We evaluate the performance of the proposed framework on a large vocabulary spontaneous conversational telephone speech dataset (Switchboard-1 Release 2). The proposed approach achieves an absolute figure of merit improvement of 3.65% over the baseline system.

A new look at discriminative training for hidden Markov models

In this paper we compare three frameworks for discriminative training of continuous-density hidden Markov models (CD-HMMs). Specifically, we compare two popular frameworks, based on conditional maximum likelihood (CML) and minimum classification error (MCE), to a new framework based on margin maximization. Unlike CML and MCE, our formulation of large margin training explicitly penalizes incorrect decodings by an amount proportional to the number of mislabeled hidden states. It also leads to a convex optimization over the parameter space of CD-HMMs, thus avoiding the problem of spurious local minima. We used discriminatively trained CD-HMMs from all three frameworks to build phonetic recognizers on the TIMIT speech corpus. The different recognizers employed exactly the same acoustic front end and hidden state space, thus enabling us to isolate the effect of different cost functions, parameterizations, and numerical optimizations. Experimentally, we find that our framework for large margin training yields significantly lower error rates than both CML and MCE training.

A method and system used to determine the similarity between an input speech data and a sample speech data is provided. First, the input speech data is segmented into a plurality of input speech frames and the sample speech data is segmented into a plurality of sample speech frames. Then, the input speech frames and the sample speech frames are used to build a matching matrix, wherein the matching matrix comprises the distance values between each of the input speech frames and each of the sample speech frames. Next, the distance values are used to calculate a matching score. Finally, the similarity between the input speech data and the sample speech data is determined according to this matching score.

Clouds, Continuums, and Cells

Although it has been studied in some depth, texture characterization is still a challenging issue for real-life applications. In this study, we propose a multiresolution salient-point-based approach in the wavelet domain. This incorporates a two-phase feature extraction scheme. In the first phase, each wavelet subband (LH, HL, or HH) is used to compute local features by using multidisciplined (statistical, geometrical, or fractal) existing texture measures. These features are converted into binary images, called salient point images (SPIs), via threshold operation. This operation is the key step in our approach because it provides an opportunity for better segmentation and combination of multiple features. In the final phase, we propose a set of new texture features, namely, salient-point density (SPD), non-salient-point density (NSPD), salient-point residual (SPR), saliency and non-saliency product (SNP), and salient-point distribution non-uniformity (SPDN). These features characterize various aspects of image texture such as fineness/coarseness, primitive distribution, internal structures, etc. These features are then applied to the well-known K-means algorithm for unsupervised segmentation of texture images. Experimental results with the standard texture (Brodatz) and natural images demonstrate the robustness and potential of the proposed features compared to the wavelet energy (WE) and local extrema density feature (LED).

Text summarization produces a concise amount of data out of document which may represent the core information contained in the document. In this paper, a combinational approach has been used to generate summary from newspaper article that uses a number of features to rank sentences of the article. Graph based approach with minimal text sentiment analysis and the presence of category wise key phrases have been used. Sentences with the sentiment polarity of the overall document sentiment polarity can convey important information. Combination of multiple features can lead to the identification process of the significant sentences with which good summary can be formed. After the performance analysis of the approach, it has been observed that the technique can generate good summaries that are close to human written ones.

In multimedia-based e-learning systems, there are strong needs for segmenting lecture videos into topic units in order to organize the videos for browsing and to provide search capability. Automatic segmentation is highly desired because of the high cost of manual segmentation. While a lot of research has been conducted on topic segmentation of transcribed spoken text, most attempts rely on domain-specific cues and formal presentation format, and require extensive training; none of these features exist in lecture videos with unscripted and spontaneous speech. In addition, lecture videos usually have few scene changes, which imply that the visual information that most video segmentation methods rely on is not available. Furthermore, even when there are scene changes, they do not match with the topic transitions. In this paper, we make use of the transcribed speech text extracted from the audio track of video to segment lecture videos into topics. We review related research and propose a segmentation approach. Our approach utilizes features such as noun phrases and combines multiple content-based and discourse-based features. Our preliminary results show that the noun phrases are salient features and the combination of multiple features is promising to improve segmentation accuracy.

Moving objects segmentation is the foundation of intelligent moving body information collection. In the monocular vision system, it is too difficult to segment the foreground from background when their grayscale and color are similar. Compared with the grayscale and color, the scenery depth are less susceptible to external environment, if depth information can be got in stereo vision, it would be much easier to segment the foreground. Unfortunately, scenery accurate dense disparity map is often hard to get. In the paper, an optimizing method of calculating the dense disparity based on graph cut is used, and based on the dense disparity, the features of color and disparity are combined by graph cut to improve the accuracy of the image segmentation. The experimental results show that the method provides a better dense disparity map and also reduces the effect of light and strengthens the stability of segmentation by combining the features of color and disparity in graph cut optimization algorithm.