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.     

Dynamics of the formation of two-dimensional ordered structures

The growth of ordered domains in lattice gas models, which occurs after the system is quenched from infinite temperature to a state below the critical temperatureT _c, is studied by Monte Carlo simulation. For a square lattice with repulsion between nearest and next-nearest neighbors, which in equilibrium exhibits fourfold degenerate (2×1) superstructures, the time-dependent energy E(t), domain size L(t), and structure functionS(q, t) are obtained, both for Glauber dynamics (no conservation law) and the case with conserved density (Kawasaki dynamics). At late times the energy excess and halfwidth of the structure factor decrease proportional tot ^–x, whileL(t) t ^x, where the exponent x=1/2 for Glauber dynamics and x1/3 for Kawasaki dynamics. In addition, the structure factor satisfies a scaling lawS(k,t)=t ^2xS(kt^x). The smaller exponent for the conserved density case is traced back to the excess density contained in the walls between ordered domains which must be redistributed during growth. Quenches toT>T _c, T=T_c (where we estimate dynamic critical exponents) andT=0 are also considered. In the latter case, the system becomes frozen in a glasslike domain pattern far from equilibrium when using Kawasaki dynamics. The generalization of our results to other lattices and structures also is briefly discussed.     

Dynamic Lie algebraic formulation of second‐order optical nonlinearity for substituted benzenes

A dynamic Lie algebraic (DLA) formulation is applied to the study of nonlinear optical properties of the substituted benzenes. We have described the generation of the dynamic Lie algebra for the model Hamiltonian used in the present study. In terms of these elements of the dynamic algebra we express the evolution operator as a function of the group parameters, which can be determined by means of solving a system of coupled nonlinear differential equations. Thus, in terms of the density matrix operator formalism in statistical mechanics we obtain the statistical averages of the electric polarization and then derive readily an expression for the hyperpolarizability of the paradisubstituted benzenes. Comparisons with experimental observations and other quantum calculations of the hyperpolarizabilities for the paradisubstituted benzenes are made qualitatively and quantitatively. These results imply that the DLA method appears useful in describing the nonlinear optical phenomena in the substituted benzene molecules. © 2003 Wiley Periodicals, Inc. Int J Quantum Chem 93: 335–343, 2003     

Study of temperature transitions in solid poly(oxymethylene) by thermal analysis methods

Poly(oxymethylene)s with different molecular mass and chemical structure were studied using DSC, dynamic mechanical relaxation, thermomechanical analysis, and thermogravimetry.Molecular mobilities of two types were found in the amorphous phase of poly(oxymethylene). Unconstrained chains of poly(oxymethylene) soften at –70°C and then amorphous chains with different restraints from the crystalline phase are successively activated in a wide temperature interval.Dedicated to Professor Bernhard Wunderlich on the occasion of his 65th birthdayThis work has been supported by DuPont Company     

On the origin of dynamic instability of molecular systems

The vibronic origin of dynamic instability of molecular systems considered earlier, is here given a more complete and rigorous treatment. It is shown that the nonvibronic contribution to the curvature of the adiabatic potential arising due to nuclear displacements under fixed electronic density distribution, is always positive, and hence the only reason for dynamic instability is the pseudo Jahn-Teller effect. For some examples of special interest (planar equilateral NH₃, planar square CH₄ and linear H ₃ ⁺ ) the molecular excited states, responsible for the instability of the ground state, are revealed by means of ab initio calculations.     

Mechanism of Cp<Subscript>2</Subscript>ZrCl<Subscript>2</Subscript>-catalyzed olefin hydroalumination by alkylalanes

The composition of intermediates of the Cp₂ZrCl₂-catalyzed hydroalumination of α-olefins by isobutylalanes (HAlBuⁱ ₂, AlBuⁱ ₃, ClAlBuⁱ ₂) was studied by dynamic ¹H and ¹³C NMR pectroscopy. The reaction of Cp₂ZrCl₂ with isobutylalanes affords the complex (Cp₂ZrHCl·HAlBuⁱ ₂)₂ responsible for α-olefin hydroalumination.__________Published in Russian in Izvestiya Akademii Nauk. Seriya Khimicheskaya, No. 2, pp. 311–322, February, 2005.     

Orthokinetic Aggregation in Two Dimensions of Monodisperse and Bidisperse Colloidal Systems

Orthokinetic aggregation of colloids trapped at the air–liquid interface was studied by direct imaging in a couette cell. This method allowed us to follow the temporal evolution of both the cluster-mass distribution and the cluster structure at a shear rate where Brownian aggregation is suppressed. The interactions between the monodisperse latex particles floating at the air–liquid interface were controlled either by varying the electrolyte concentration or by creating a bidisperse system through the addition of small particles. The results show that the clusters in all of the systems are characterized by a high fractal dimension, indicating that the clusters are rearranged and densified by the shear. Kinetic analysis suggests that aggregation of monodisperse systems mainly proceeds through homogeneous aggregation, i.e., large clusters sticking to other large clusters. The bidisperse system, finally, with a size ratio around 10, favored a more heterogeneous aggregation among small and large clusters throughout the aggregation process; a slightly lower fractal dimension was observed compared to the strongly aggregated monodisperse system.     

Dynamic covalent approach to [2]- and [3]roxtanes by utilizing a reversible thiol-disulfide interchange reaction

A dynamic covalent approach to disulfide-containing [2]- and [3]rotaxanes is described. Symmetrical dumbbell-shaped compounds with two secondary ammonium centers and a central located disulfide bond were synthesized as components of rotaxanes. The rotaxanes were synthesized from the dumbbell-shaped compounds and dibenzo-[24]crown-8 (DB24C8) with catalysis by benzenethiol. The yields of isolated rotaxanes reached about 90 % under optimized conditions. A kinetic study on the reaction forming [2]rotaxane 2 a and [3]rotaxane 3 a suggested a plausible reaction mechanism comprising several steps, including 1) initiation, 2) [2]rotaxane formation, and 3) [3]rotaxane formation. The whole reaction was found to be reversible in the presence of thiols, and thermodynamic control over product distribution was thus possible by varying the temperature, solvent, initial ratio of substrates, and concentration. The steric bulk of the end-capping groups had almost no influence on rotaxane yields, but the structure of the thiol was crucial for reaction rates. Amines and phosphines were also effective as catalysts. The structural characterization of the rotaxanes included an X-ray crystallographic study on [3]rotaxane 3 a.     

Rapid and sensitive determination of ephedrine and pseudoephedrine by micellar electrokinetic chromatography with on-line regenerating covalent coating

A rapid, sensitive and reproducible micellar electrokinetic chromatographic method using hexamethyldisilazane as on-line regenerating covalent coating was developed for the quantification of ephedrine (E) and pseudoephedrine (PE). E and PE were derivatized with 4-chloro-7-nitrobenzo-2-oxa-1,3-diazol for laser-induced fluorescence detection. The on-line regenerating covalent coating formed a combinative double coating with the subsequently produced dynamic SDS coating. The total coating can be easily removed and conveniently regenerated on-line. The simple coating procedure was described. By a series of optimization, a running buffer of 20 mm Na(2)B(4)O(7) + 16 mm SDS was applied for the separation of the derivatives. Linear relationships for E and PE were obtained in the range of 0.044-6.60 microg mL(-1) (correlation coefficients: 0.9975 for E, 0.9981 for PE), and the detection limits for E and PE were 1.71 and 0.67 ng mL(-1), respectively. The separation speed, the reproducibility and the sensitivity were much improved over those of other capillary electrophoresis methods more recently reported. The method was applied to the analysis of the two alkaloids in traditional herbal preparations with recoveries in the range 92.8-104.8%.     

Fast and reversible migrations of N,S-centered groups around the perimeter of cyclopropene and cycloheptatriene rings

The kinetics and mechanism of circumambulatory rearrangements of N-centered (NCS) and S-centered (SPh, SC₃Ph₃, SC(OEt)=S) groups in corresponding derivatives of 1,2,3-triphenylcyclopropene and cycloheptatriene were studied by dynamic¹H and¹³C NMR spectroscopy. Migrations of the isothiocyanate group occur by the dissociation-recombination mechanism with intermediate formation of a tight ionic pair. Migrations of the phenylthio group around the perimeter of cyclopropene and cycloheptatriene rings occur by the 1,2-shift mechanism. It was found that rearrangements of theO-ethyl dithiocarbonate group inS-(1,2,3-triphenylcyclopropen-3-yl)-O-ethyl dithiocarbonate occur by the 3,3-sigmatropic shift mechanism. The molecular and crystal structure ofO-ethylS-(1,2,3-triphenylcyclopropen-3-yl) dithiocarbonate was studied by X-ray analysis.