This paper addresses the extraction of multipurpose spectral rhythm features that simultaneously tackle a variety of rhythm analysis tasks, namely, dance style classification, meter estimation, and tempo estimation. The term spectral rhythm features emanates from the origin of the extracted features, which is the periodicity function (PF), a spectral representation that encapsulates the salience of the rhythm frequencies. Two dimensionality reduction techniques applied on the PF to extract expressive and compact features are compared, namely, a linear transformation resulting from Principal Component Analysis and a nonlinear mapping derived from a Restricted Boltzmann Machine. Subsequently, the derived features were used as input to an SVM classifier for each task. Moreover, an additional method is proposed that reformulates the well-studied tempo estimation task as a combination of multiple binary classification sub-problems. Evaluation was performed on a large number of datasets demonstrating that the same set of features learned from the PF provide a robust rhythmic representation that achieved comparable results to the current state-of-the-art methods for the aforementioned tasks.

A critical issue in recognition of mathematical expressions is the identification of the spatial relations of the symbols or/and sub-expressions that comprise the entire mathematical formula. This paper addresses the problem of structural analysis of mathematical expressions by constructing appropriate feature vectors to represent the spatial affinity of the objects (mathematical symbols or sub-expressions) under examination and employing two popular machine learning techniques: (i) Support Vector Machines (SVM) and (ii) Artificial Neural Networks (ANN) to recognize the spatial relation between these objects. In order to evaluate the proposed techniques, we use Math Brush, a large publicly available dataset of mathematical expressions with annotated spatial relations, and a subset of spatial relations derived from the mathematical expressions the CROHME 2012 dataset. The experimental results give an overall mean error rate of 2.8% for the SVM and 3.4% for the ANN classifiers respectively, which are at par with other approaches evaluated on the same datasets.

In this paper, we present tempo estimation and beat tracking algorithms by utilizing percussive/harmonic separation of the audio signal, in order to extract filterbank energies and chroma features from the respective components. Periodicity analysis is carried out by the convolution of feature sequences with a bank of resonators. Target tempo is estimated from the resulting periodicity vector by incorporating metrical relations knowledge. Tempo estimation is followed by a local tempo refinement method to enhance the beat-tracking algorithm. Beat tracking involves the computation of the beat saliencies derived from the resonators responses and proposes a distance measure between candidate beats locations. A dynamic programming algorithm is adopted to find the optimal “path” of beats. Both tempo estimation and beat tracking methods were submitted on MIREX 2011, while the tempo estimation algorithm was also evaluated on ISMIR 2004 Tempo Induction Evaluation Exchange Dataset.

Document image binarization is an initial though critical stage towards the recognition of the text components of a document. This paper describes an efficient method based on mathematical morphology for extracting text regions from degraded handwritten document images. The basic stages of our approach are: (a) top-hat-by-reconstruction to produce a filtered image with reasonable even background, (b) region growing starting from a set of seed points and attaching to each seed similar intensity neighboring pixels and (c) conditional extension of the initially detected text regions based on the values of the second derivative of the filtered image. The method was evaluated on the benchmarking dataset of the International Document Image Binarization Contest (DIBCO 2011) and show promising results.

We present a system for recognizing online mathematical expressions (ME). Symbol recognition is based on a template elastic matching distance between pen direction features. The structural analysis of the ME is based on extracting the baseline of the ME and then classifying symbols into levels above and below the baseline. The symbols are then sequentially analyzed using six spatial relations and a respective 2d structure is processed to give the resulting MathML representation of the ME. The system was evaluated on the Competition on Recognition of Online Handwritten Mathematical Expressions (CROHME) 2011 datasets and demonstrates promising results.

In this paper, the use of closed-form expressions is compared to the BIC approximation, with respect to speaker clustering. We first show that the particular BIC setting which is commonly used in this task, namely the approximation of the marginal with respect to the model parameters and conditional with respect to the latent variables likelihood, belongs to an exponential family, and hence admits a closed-form expression by attaching conjugate priors. We then formalize the role of the tuning parameter as a hyperparameter of the prior and finally we explain the several proposed setting global, local and segmental based on the strength of the prior. Experiments are carried out for the speaker clustering task and improvement over the BIC approximation is reported.

In this paper we examine a new penalty term for the Bayesian Information Criterion (BIC) that is suited to the problem of speaker diarization. Based on our previous approach of penalizing each cluster only with its effective sample size - an approach we called segmental - we propose a stricter penalty term. The criterion we derive retains the main property of the Segmental-BIC, i.e. it approximates the evidence of overall partitions of the data and simultaneously leads to a pairwise dissimilarity measure that is completely defined by the pair of clusters in question. The experimental results show significant improvement in diarization accuracy on the ESTER benchmark.

Document image segmentation to text lines is a critical stage towards unconstrained handwritten document recognition. Although morphological operations proved to be effective in processing machine-printed documents for several issues, similar methods for unconstraint-handwritten documents lack accuracy. We propose an efficient method based on binary morphology for text-line segmentation in such documents. The basic steps of our approach are: a) sub sampling and binary rank order filtering to enhance the text-line structures and b) applying dilations and (p,q)-th generalized foreground rank openings successively to join close and horizontally overlapping regions while preventing a merge in the vertical direction. The method tested on the benchmarking dataset of the ICDAR07 handwriting segmentation contest and show remarkable results.

A novel approach to the Bayesian Information Criterion (BIC) is introduced. The new criterion redefines the penalty terms of the BIC, such that each parameter is penalized with the effective sample size is trained with. Contrary to Local-BIC, the proposed criterion scores overall clustering hypotheses and therefore is not restricted to hierarchical clustering algorithms. Contrary to Global-BIC, it provides a local dissimilarity measure that depends only the statistics of the examined clusters and not on the overall sample size. We tested our criterion with two benchmark tests and found significant improvement in performance in the speaker diarisation task. Copyright © 2009 ISCA.

This paper addresses the problem of automatic text-line and word segmentation in handwritten document images. Two novel approaches are presented, one for each task. In text-line segmentation a Viterbi algorithm is proposed while an SVM-based metric is adopted to locate words in each text-line. The overall algorithm was tested in the ICDAR2007 handwriting segmentation contest and showed highly promising results.

In this paper we present a method of combining several acoustic parametric spaces, statistical models and distance metrics in speaker diarization task. Focusing our interest on the post-segmentation part of the problem, we adopt an incremental feature selection and fusion algorithm based on the Maximum Entropy Principle and Iterative Scaling Algorithm that combines several statistical distance measures on speech-chunk pairs. By this approach, we place the merging-of-chunks clustering process into a probabilistic framework. We also propose a decomposition of the input space according to gender, recording conditions and chunk lengths. The algorithm produced highly competitive results compared to GMM-UBM state-of-the-art methods.

A new method for verifying text areas detected in video streams is proposed. The algorithm explores the spectral properties of the horizontal projection of candidate text regions in order to reduce the high amount of false alarms that most text detection algorithms suffer from. The full algorithm (text localization followed by verification and temporal redundancy module) has been tested on newscast video sequences (MPEG-1-720 x 576 resolution-184 minutes). The detection module produced 94.82% recall rate but only 51.84% precision rate. The addition of the verification module increased the precision rate to 78.93% keeping the recall rate almost unaffected.

We report on the application of the Self-Organizing Map (SOM) classification method to the task of categorizing texts according to their register and the style of their author. The SOM has been selected as its performance in various data-mining applications has been found to be highly successful. Here, the method is evaluated against the task of clustering textual data which are corpora of texts written in the Greek language; the parameters used depict linguistically important structural properties of the texts. The experiments reported indicate that the SOM results are equivalent to those generated by statistical methods.

A new state-space method for spectral estimation that performs decimation by factor two while it makes use of the full set of data available is presented. The proposed method, called DESE2, is based on singular value decomposition in order to estimate frequency, damping factor, amplitude and phase of exponentially damped sinusoids in the presence of noise. The DESE2 method is compared against some previously proposed methods for spectral estimation that lie among the most promising methods in the field of spectroscopy, where accuracy of parameter estimation is of utmost importance. Experiments performed on a typical simulated NMR signal prove the new method to be more robust, especially for low signal to noise ratio. The new method outperforms the other two not only by presenting lower failure rates but also by incorporating enhanced discriminative analysis while at the same time it benefits from the use of the full data set.

Abstract In this article, a methodology is presented regarding thedesign of a segment database for use with a time-domainspeech synthesis system for the Greek language. The mainissue of this process is the systematic generation of a corpuscontaining all possible instances of the segments for thespecific language. Particular issues such as the phoneticcoverage, the sentence selection as well as iterative evaluationtechniques employing custom-built tools are discussed. Theresulting corpus is characterised by a near-minimal size,provides a complete coverage of the Greek language and itsdistribution of phonemes is similar to that of natural corpora.A typical spoken acquisition procedure may then beperformed, resulting in a segment database for use with atime-domain Greek synthesizer. The corpus creationprocedure allows for the fine-tuning of the segment database’slanguage-dependent characteristics and thus assists in thegeneration of high-quality text-to-speech synthesis. 1. Introduction

This paper presents work performed for the Time-Domain TTS system, which is being developed at the ILSP for the Greek language. It focuses on the enhancement of the synthetic speech quality, by reducing the spectral mismatches between concatenated segments. To that end, a study has been performed to determine the distance that can best predict when a spectral mismatch is audible. Experimentation with different spectral distances has taken place and the distance with the best performance has been used in order to systematically enrich the segment database, which initially contained only one instance per segment. Results of this procedure indicate a substantial improvement on the synthetic speech quality.

This paper presents a new state-space method for spectral estimation based on a companion matrix technique in order to estimate frequency, damping factor, amplitude and phase of exponential sinusoids. The new method, called CSE (companion matrix based spectral estimation), is compared against a previously proposed method called HTLS which is based on the use of total least squares. The latter lies among the most promising methods in the field of spectroscopy where accuracy of parameter estimation is of utmost importance. Experiments performed on a simulated NMR signal prove CSE to be more robust in terms of failure rate, especially for low signal to noise ratio.

Fuzzy set theory and fuzzy logic has been initiated by Zadeh back in 1965 to permit the treatment of vague, imprecise, and ill-defined knowledge in an concise manner. One of the unique advantages of fuzzy logic is that it is capable of directly incorporating and utilizing qualitative and heuristic knowledge in the form of causal if-then production rules for reasoning and inference. On the other hand, rule-based speech synthesis based on formants makes considerable use of rules for numerous of the tasks it involves, e.g. graphemic to phonemic transcription, coarticulation, concatenation, and duration rules etc. These rules also take the if-then form with their antecedent (condition) part describing the context of the rule and their decedent an appropriate action to be taken. The main motivation for introducing fuzzy logic in the synthesis-by-rule paradigm, is its ability to host and treat uncertainty and imprecision both in the condition part of the rule as well as its decedent part. This may be argued to significantly reduce the number of required rules while rendering them more meaningful and human-like.

The frequency analysis of a signal is considered. Although the classical approach based on the representation of a signal in terms of complex exponentials has its merit, an alternative approach aiming to represent a signal by a sum of exponential functions is also interesting. An exact solution to this problem is found at the meeting point of linear predictive coding (LPC) and modeling techniques based on the singular value decomposition (SVD) analysis. Further insight in the structural analysis of a signal is provided. The rank of a signal is defined, and it is shown that full recovery of finite rank data from an incomplete set is always possible.>

The error propagation in various useful order- and time-recursive DSP algorithms is studied. It is demonstrated that in all these algorithms there are two kinds of error due to finite precision: an erratic and a systematic one. Examples of both kinds of error are provided, and special emphasis is given to the study of the systematic truncation or round-off error. It is shown that in the Levinson-type and Schur-type algorithms for the solution of the FLP (forward linear prediction) and the FIR (finite impulse response) problem, there is one dominant source of systematic error, while in the l-step ahead case there are two sources of such error. Moreover, it is pointed out that in the fast Kalman algorithms there are two kinds of systematic error. Precise indicators of the exact magnitude of the finite precision error are given, and possible recovery techniques are proposed.>

The numerical behavior of the order-recursive algorithms for the solution of a number of typical Toeplitz systems is studied. Use is made of a novel methodology that gives a direct interpretation of the positive definiteness that leads to the explicit computation of the minimum absolute bounds of the 1-step and l-step-ahead linear predictors. Ill-conditioning is discussed, and an algorithmic scheme with improved numerical behavior is given for the l-step-ahead LP case. It is proved that the Levinson algorithm, when it is used for the solution of the least-squares finite-impulse-response filtering problem, manifests a fundamentally different numerical behavior than when it is used for the l-step-ahead problem.>

A fast computationally efficient method for automatic segmentation and classification of textured images is presented. The method does not necessarily need a-priori information about the textures present in the image, thus avoiding the necessity of a training set of textures. A fast adaptive multichannel technique for autoregressive image model parameter estimation with fast tracking capabilities and a powerful statistical distance measure are appropriately interweaved to form the proposed technique. Specific properties of the estimation part of the algorithm are exploited to reduce greatly the computational complexity of the distance measure. Some interesting extensions of the method are discussed and examples are given which illustrate the performance of the algorithm.>

An autocorrelation measure of a single-frame which is not derived directly from the data is defined. A lattice method is first used to compute the frame-reflection coefficients. Then the predictor coefficients are computed, and the normal equations result in a new set of autocorrelation lags named the lattice autocorrelation measure (LAM). LAM is shown to be more accurate than the conventional Bartlett autocorrelation measure, especially in case of few samples and low SNR. The application of LAM in eigenvalues spectral analysis gives better estimates than conventional autocorrelation, mainly in the central frequency range. The use of the LAM in overdetermined autocorrelation matrices analyzed by singular value decomposition leads to an improvement in the frequency estimation of very close harmonics in the cases of short signal frames.>

A fast, computationally efficient method for adaptive image estimation is presented in this paper. The method is based on the multichannel form of the Fast-A-Posteriori-Error Sequential Technique (FAEST) for the estimation of the parameters of 2-D autoregressive (AR) models. The above models are appropriately designed to have the shift invariance property necessary for fast recursive least squares techniques. Moreover inclusion of a forgetting factor, or of a sliding window in the algorithm, permits adaptive image model parameter estimation. The specific properties of the algorithm are examined and it is shown that the proposed estimation scheme is much faster than other existing approaches. Various interesting applications of the method are discussed and examples are given which illustrate the above theoretical results.

This paper presents order-recursive algorithms for the computation of the model parameters required in Multipulse Linear Predictive Coding (LPC) of speech in the case of autocorrelation and covariance methods. Its computational complexity is 2p2+ 4p multilications and divisions, where p+1 is the number of the model parameters. It is also shown, that the developed algorithm leads in a natural way, to lattice-ladder structures for multipulse LPC analysis of speech.

The Shür algorithm is considered in this paper and two different computational organizations are studied. The first is equivalent to the well known "parallel" solution of the Toeplitz equations, the second gives an order recursive algorithm. A simple demonstration of the Shür recursion is outlined and a "superlattice" structure is given for the imple - mentation of both types of algorithms.

The purpose of this paper is to provide a brief overview of the algorithms and structures which can be used in connection with ℓ - lag FIR filtering. A number of existing efficient algorithms from the zero - lag problem are generalized here, to be applied to the ℓ - lag case. Both direct and lattice - ladder structures are considered. Four different families of techniques are discussed: time recursive, order recursive, lattice - ladder and "recursive in lag". The unified approach used in this paper for the derivation of the algorithms permits the underlining of the relationships among the variables appearing in the four families of recursive schemes discussed. The "prewindowing" assumption is used throughout.

Sequential Least-Squares (LS) methods play a prominent role in many digital signal processing applications. The conventional implementation of these schemes requires an amount of operations proportional to the square of the number of estimated parameters. In contrast a variety of existing fast algorithms offer a computational complexity proportional to the number of estimated parameters. Such schemes exist for both direct form and lattice-ladder filter structures. This paper offers a unified overview of fast sequential algorithms for LS FIR filters, implemented using a direct form realization, in the case of prewindowed multichannel signals. Although all these algorithms are theoretically equivalent, in practice they exhibit different performance due to round-off noise, incorrect initialization etc. The performance evaluation of all these shemes is still an area to be explored.

The present paper deals with a new, computationally efficient, algorithm for Sequential Least Squares (LS) estimation. This scheme requires only O(5p) MAD (Multiplications And Divisions) per recursion to update a Kalman type gain vector; p is the number of estimated parameters. In contrast the well-known fast Kalman algorithm requires O(8p) MAD. The introduced method is the fastest known algorithm featured by the rapid convergence characteristics of exact Least Squares estimation schemes. Another interesting feature of the new algorithm is the balanced role, the forward and backward prediction play.

In various signal processing applications we are confronted with the problem of finding the optimum in the least squares sense linear filter that matches the input signal with a certain lag of the desired response. The system of equations yielding all filters coefficients turns out to have a near-to-Toeplitz or near-to-Hankel structure with displacement rank 1 or 2 generically depending on whether windowing is used. Motivated by these questions the general linear system is considered and recursive procedures are derived for the determination of the solution which when the parameters have low displacement rank are proved very efficient. In particular if the parameters are Toeplitz the resulting algorithm coincides with a scheme introduced recently by the authors which in turn performs better than the well-known Simpson's sideways recursions.

This paper introduces a sequential fast scheme of the Kalman type to solve the covariance equations of linear prediction. The new algorithm constitutes a generalization of the scheme introduced recently by D. Falconer and L. Ljung, and applies to a more general situation. The algo rithm is characterized by conceptual simplicity and efficiency. Although it is based on matrix partitioning concepts and shifting properties of the signal vector, it does not have the restriction of prewindowing as it is the case with the previously mentioned algorithm, thus it can be used to model small frames of signals as well as to model signals generated by an AR model, giving in this case the exact solution. Finally it must be noted that the algorithm is strictly time recursive and leads to O(13p) multiplications per recursion only. Thus, this new scheme is the fastest existing method for the sequential solution of the general covariance equations when prewindowing is not desirable.

In various signal processing applications one is often confronted with aspects such as linear system solution, triangularization or inversion of matrices with special block structure as well as entries of particular form. Toeplitz, Banded Toeplitz, circular and Hankel matrices provide typical examples often encountered in such diverse fields as image processing, computerized tomography and other array processing applications. The purpose of this paper is to algorithmically examine the issues of triangularization, inversion and linear system solution when the above particular structures are imposed at either the block level or the entry level. It is shown that the various resulting combinations of block and entry structure considerably reduce the computational complexity of the above problems.

The purpose of this paper is to give reduced complexity algorithms in order to compute linear models in situations of non-stationnary signals. The case of impulse response identification is examined as well as the case of pure AR-modeling. Both problems lead to a linear system which must be solved to give the unknown AR-model or the unknown impulse response. Furthermore the linear systems that occur in both cases have an identical left hand side member characterized by a symmetric matrix having the supplementary property of being the product of two non-symmetric Toeplitz matrices. The "near to Toeplitz" structure of this matrix permits the fast solution of the system with specially adapted algorithms to this situation. Many properties of the near to Toeplitz matrices are outlined as the key formulas for the derivation of all fast schemes. Order, time and order-time recursive methods are derived using a unified theoretical framework. The methodology used here is similar to that used by Morf et al. The algorithms presented are new and can be converted to efficient and automated computer packages.

This paper introduces a modified linear prediction method based on the Karhumen-Loeve expansion of the correlation matrix of the speech samples. This result is obtained via a new normalization of the parameters. It is shown that, due to some important properties of Toeplitz matrices, the poles of the A. R. model lie on the unit circle. Consequently, only the formant frequencies are computed and the result can be interpreted as a special discrete Fourier transform. Application to speech analysis is developped with a comparison to the usual linear prediction and ceptrum methods.