Aligning Text and Phonemes for Speech Technology Applications Using an EM-Like Algorithm

A common requirement in speech technology is to align two different symbolic representations of the same linguistic ‘message’. For instance, we often need to align letters of words listed in a dictionary with the corresponding phonemes specifying their pronunciation. As dictionaries become ever bigger, manual alignment becomes less and less tenable yet automatic alignment is a hard problem for a language like English. In this paper, we describe the use of a form of the expectation-maximization (EM) algorithm to learn alignments of English text and phonemes, starting from a variety of initializations. We use the British English Example Pronunciation (BEEP) dictionary of almost 200,000 words in this work. The quality of alignment is difficult to determine quantitatively since no ‘gold standard’ correct alignment exists. We evaluate the success of our algorithm indirectly from the performance of a pronunciation by analogy system using the aligned dictionary data as a knowledge base for inferring pronunciations. We find excellent performance—the best so far reported in the literature. There is very little dependence on the start point for alignment, indicating that the EM search space is strongly convex. Since the aligned BEEP dictionary is a potentially valuable resource, it is made freely available for research use.     

Composite Likelihood Inference in a Discrete Latent Variable Model for Two-Way “Clustering-by-Segmentation” Problems

We consider a discrete latent variable model for two-way data arrays, which allows one to simultaneously produce clusters along one of the data dimensions (e.g., exchangeable observational units or features) and contiguous groups, or segments, along the other (e.g., consecutively ordered times or locations). The model relies on a hidden Markov structure but, given its complexity, cannot be estimated by full maximum likelihood. Therefore, we introduce a composite likelihood methodology based on considering different subsets of the data. The proposed approach is illustrated by simulation, and with an application to genomic data.     

Selecting hidden Markov model state number with cross-validated likelihood

The problem of estimating the number of hidden states in a hidden Markov model is considered. Emphasis is placed on cross-validated likelihood criteria. Using cross-validation to assess the number of hidden states allows to circumvent the well-documented technical difficulties of the order identification problem in mixture models. Moreover, in a predictive perspective, it does not require that the sampling distribution belongs to one of the models in competition. However, computing cross-validated likelihood for hidden Markov models for which only one training sample is available, involves difficulties since the data are not independent. Two approaches are proposed to compute cross-validated likelihood for a hidden Markov model. The first one consists of using a deterministic half-sampling procedure, and the second one consists of an adaptation of the EM algorithm for hidden Markov models, to take into account randomly missing values induced by cross-validation. Numerical experiments on both simulated and real data sets compare different versions of cross-validated likelihood criterion and penalised likelihood criteria, including BIC and a penalised marginal likelihood criterion. Those numerical experiments highlight a promising behaviour of the deterministic half-sampling criterion.     

A VARIATIONAL EXPECTATION-MAXIMIZATION METHOD FOR THE INVERSE BLACK BODY RADIATION PROBLEM

The inverse black body radiation problem, which is to reconstruct the area temperature distribution from the measurement of power spectrum distribution, is a well-known ill-posed problem. In this paper, a variational expectation-maximization （EM） method is developed and its convergence is studied. Numerical experiments demonstrate that the variational EM method is more efficient and accurate than the traditional methods, including the Tikhonov regularization method, the Landweber method and the conjugate gradient method.     

Attenuation properties of EM wave by magnetized nonuniform plasma slab coated on perfect conductor plane

The total amplitude reflection coefficient of EM wave in a nonuniform plasma slab coated on perfect conductor plane is newly derived by using the scattering matrix method (SMM), and the attenuation is calculated. Three types of plasma electron density profile, that is the hyperbolic, sinusoidal, and linear profiles, are used. The external magnetic field and plasma parameters such as the maximum electron density and collision frequencies are discussed to calculate the attenuation of EM wave. The calculation results show that the plasma electron density profile and maximum electron density exert a large effect to the attenuation of EM wave and the attenuation under the uniform external magnetic field is taken place by the electron cyclotron resonance absorption, the up hybrid resonance absorption and geometric resonances absorption.     

A comparative study of dispersion relation of EM waves in ternary one-dimensional plasma photonic crystals having two different structures

In this paper, a rigorous theoretical analysis has been made to study the dispersion relation of EM waves in periodic ternary one-dimensional photonic crystal having two different structures. In one case we have chosen glass-plasma and ZnS in one unit cell and in other case we have considered glass-plasma and M_gF₂ in one unit cell. Using Kronig-Penney model the dispersion relation for proposed structures has been obtained and numerical results are presented in the form of dispersion curves. The dependence of photonic band gap (PBG) characteristics on plasma frequency, plasma width and the width of dielectric media are discussed in the light of frequency gap and cutoffs of binary one-dimensional plasma photonic crystal. An attempt has been made to show how the PBG characteristic of a particular structure changes when the dielectric materials of its unit cell is changed by the other dielectric material. It is found that the structure having glass-plasma-ZnS in unit cell is more useful for broad band filtering and other plasma functioning devices compared to the structure having glass-plasma-MgF₂ in one unit cell.     

The use of trehalose in the preparation of specimens for molecular electron microscopy

Biological specimens have to be prepared for imaging in the electron microscope in a way that preserves their native structure. Two-dimensional (2D) protein crystals to be analyzed by electron crystallography are best preserved by sugar embedding. One of the sugars often used to embed 2D crystals is trehalose, a disaccharide used by many organisms for protection against stress conditions. Sugars such as trehalose can also be added to negative staining solutions used to prepare proteins and macromolecular complexes for structural studies by single-particle electron microscopy (EM). In this review, we describe trehalose and its characteristics that make it so well suited for preparation of EM specimens and we review specimen preparation methods with a focus on the use of trehalose.     

On filtering and estimation of a threshold stochastic volatility model

We derive a nonlinear filter and the corresponding filter-based estimates for a threshold autoregressive stochastic volatility (TARSV) model. Using the technique of a reference probability measure, we derive a nonlinear filter for the hidden volatility and related quantities. The filter-based estimates for the unknown parameters are then obtained from the EM algorithm.