MLLP-VRAIN Spanish ASR Systems for the Albayzín-RTVE 2020 Speech-To-Text Challenge: Extension

This paper describes the automatic speech recognition (ASR) systems built by the MLLP-VRAIN research group of Universitat Politècnica de València for the Albayzín-RTVE 2020 Speech-to-Text Challenge, and includes an extension of the work consisting of building and evaluating equivalent systems under the closed data conditions from the 2018 challenge. The primary system (p-streaming_1500ms_nlt) was a hybrid ASR system using streaming one-pass decoding with a context window of 1.5 seconds. This system achieved 16.0% WER on the test-2020 set. We also submitted three contrastive systems. From these, we highlight the system c2-streaming_600ms_t which, following a similar configuration as the primary system with a smaller context window of 0.6 s, scored 16.9% WER points on the same test set, with a measured empirical latency of 0.81 ± 0.09 s (mean ± stdev). That is, we obtained state-of-the-art latencies for high-quality automatic live captioning with a small WER degradation of 6% relative. As an extension, the equivalent closed-condition systems obtained 23.3% WER and 23.5% WER, respectively. When evaluated with an unconstrained language model, we obtained 19.9% WER and 20.4% WER; i.e., not far behind the top-performing systems with only 5% of the full acoustic data and with the extra ability of being streaming-capable. Indeed, all of these streaming systems could be put into production environments for automatic captioning of live media streams.

On Singularities of Certain Non-linear Second-Order Ordinary Differential Equations

The method of blowing up points of indeterminacy of certain systems of two ordinary differential equations is applied to obtain information about the singularity structure of the solutions of the corresponding non-linear differential equations. We first deal with the so-called Painlevé example, which passes the Painlevé test, but the solutions have more complicated singularities. Resolving base points in the equivalent system of equations we can explain the complicated structure of singularities of the original equation. The Smith example has a solution with non-isolated singularity, which is an accumulation point of algebraic singularities. Smith’s equation can be written as a system in two ways. We show that the sequence of blow-ups for both systems can be infinite. Another example that we consider is the Painlevé-Ince equation. When the usual Painlevé analysis is applied, it possesses both positive and negative resonances. We show that for three equivalent systems there is an infinite sequence of blow-ups and another one that terminates, which further gives a Laurent expansion of the solution around a movable pole. Moreover, for one system it is even possible to obtain the general solution after a sequence of blow-ups.

Equivalency in joint signatures for binary/multi-state systems of different sizes

The joint signatures of binary-state and multi-state (semi-coherent or mixed) systems with i.i.d. (independent and identically distributed) binary-state components are considered in this work. For the comparison of pairs of binary-state systems of different sizes, transformation formulas of their joint signatures are derived by using the concept of equivalent systems and a generalized triangle rule for order statistics. Similarly, for facilitating the comparison of pairs of multi-state systems of different sizes, transformation formulas of their multi-state joint signatures are also derived. Some examples are finally presented to illustrate and to verify the theoretical results established here.

Modular Neurodynamics and Its Classification by Synchronization Cores

It is assumed that the cause of cognitive and behavioral capacities of living systems is to be found in the complex structure-function relationship of their brains; a property that is still difficult to decipher. Based on a neurodynamics approach to embodied cognition this paper introduces a method to guide the development of modular neural systems into the direction of enhanced cognitive abilities. It uses formally the synchronization of subnetworks to split the dynamics of coupled systems into synchronized and asynchronous components. The concept of a synchronization core is introduced to represent a whole family of parameterized neurodynamical systems living in a synchronization manifold. It is used to identify those coupled systems having a rich spectrum of dynamical properties. Special coupling structures—called generative—are identified which allow to make the synchronized dynamics more “complex” than the dynamics of the isolated parts. Furthermore, a criterion for coupling structures is given which, in addition to the synchronized dynamics, allows also for an asynchronous dynamics by destabilizing the synchronization manifold. The large class of synchronization equivalent systems contains networks with very different coupling structures and weights allsharing the same dynamical properties. To demonstrate the method a simple example is discussed in detail.

Partial aperture imaging system based on sparse point spread holograms and nonlinear cross-correlations

Partial aperture imaging system (PAIS) is a recently developed concept in which the traditional disc-shaped aperture is replaced by an aperture with a much smaller area and yet its imaging capabilities are comparable to the full aperture systems. Recently PAIS was demonstrated as an indirect incoherent digital three-dimensional imaging technique. Later it was successfully implemented in the study of the synthetic marginal aperture with revolving telescopes (SMART) to provide superresolution with subaperture area that was less than one percent of the area of the full synthetic disc-shaped aperture. In the study of SMART, the concept of PAIS was tested by placing eight coded phase reflectors along the boundary of the full synthetic aperture. In the current study, various improvements of PAIS are tested and its performance is compared with the other equivalent systems. Among the structural changes, we test ring-shaped eight coded phase subapertures with the same area as of the previous circular subapertures, distributed along the boundary of the full disc-shaped aperture. Another change in the current system is the use of coded phase mask with a point response of a sparse dot pattern. The third change is in the reconstruction process in which a nonlinear correlation with optimal parameters is implemented. With the improved image quality, the modified-PAIS can save weight and cost of imaging devices in general and of space telescopes in particular. Experimental results with reflective objects show that the concept of coded aperture extends the limits of classical imaging.

First-order evolution equations with dynamic boundary conditions

In this paper, we introduce a general framework to study linear first-order evolution equations on a Banach space X with dynamic boundary conditions, that is with boundary conditions containing time derivatives. Our method is based on the existence of an abstract Dirichlet operator and yields finally to equivalent systems of two simpler independent equations. In particular, we are led to an abstract Cauchy problem governed by an abstract Dirichlet-to-Neumann operator on the boundary space ∂ X . Our approach is illustrated by several examples and various generalizations are indicated. This article is part of the theme issue ‘Semigroup applications everywhere’.

Dulac – Cherkas functions for systems equivalent to the van der Pol equation

The object of this study is an autonomous van der Pol system on a real plane. The subject of the study is the properties of the limit cycle of this system. The main purpose of this paper is to find the localization of the limit cycle on the phase plane and establish its shape for various values of the real parameter of the van der Pol system. Our approach is based on the use of transverse curves related to the Dulac – Cherkas functions and approximating the location of the limit cycle. As the first step, five topologically equivalent systems, including systems with a parameter rotating the vector field, as well as singularly perturbed systems are determined for the van der Pol system. Then, applying the previously elaborated method, we constructed two polynomial Dulac – Cherkas functions for each of three systems from the considered ones in the phase plane for all real nonzero values of the parameter. Using them, transverse curves forming the boundaries of the localization regions of the limit cycle for the van der Pol system are found. Thus, the constructed Dulac – Cherkas functions allow us to determine the location of the limit cycle on the basis of algebraic curves for all real parameter values, including values close to the bifurcation of a limit cycle from the center ovals, the Andronov – Hopf bifurcation, and the bifurcation from a closed trajectory related to a discontinuous periodic solution.

Modeling of Repairable Multi-State Systems with time-varying failure and repair rates using Equivalent Systems Dynamics Models

This paper treats with the reliability assessment of a Repairable Multi-State System (RMSS) by means of a Nonhomogeneous Continuous-Time Markov Chain (NH-CTMC). A RMSS run on different operating conditions that may be considered acceptable or unacceptable according to a defined demand level. In these cases, the commonly used technique is Homogeneous Continuous-Time Markov Chain (H-CTMC), since its solution is mathematically tractable. However, the H-CMTC involve that the time between state transitions is exponentially distributed, and the failure and repair rates are constants. It's certainly not true if the system components age with the operation or if the repair activities depend on the instant of time when the failure occurred. In these cases, the failure and repair rates are time-varying and the NH-CTMC is needed to be considered. Nevertheless, for these models the analytical solution may not exist and the use of others techniques is required. This paper proposes the use of an Equivalent Systems Dynamics Model (ESDM) to model a NH-CTMC. A ESDM represent the Markov Model (MM) by means of the language and the tools of the Systems Dynamics (SD), and the results are obtained by simulation. As an example, an RMSS with three components, failure rates associated with the Weibull distribution and repair rates associated with the Log-logistic distribution is developed. This example serves to identify the advantages and disadvantages of a ESDM to make model a RMSS and evaluate some reliability measures.

Parameter identification of discrete stochastic systems by the inverse iteration method

Получены условия глобальной сходимости алгоритмов, основанных на обратных итерациях в переменной метрике, в задаче идентификации параметров дискретной стохастической системы с возмущениями в невязке уравнения и наблюдениях процессов. Доказана сходимость оценок параметров к истинному значению при увеличении объема выборки наблюдений истинного процесса. Приведены примеры расчетов The article addresses the problem of identifying parameters of discrete stochastic systems with perturbations in the residual of the equation and observation of variables. The identification functional in the problem has a complex nature of isosurfaces, which is why universal minimization algorithms based on estimates of the first and second derivatives have a small radius of convergence. It is proposed to employ efficient computational identification algorithms with inverse iterations in a variable metric for solving the convergence problem for two classes of systems with simple correspondence between matrix elements and parameters of equivalent systems without state variables. These algorithms are used for systems without state variables due to the large radius and high convergence rate since the 1970s. At first, a theorem on the conditions for convergence of inverse iterations from almost any initial approximation to a small neighborhood of the global minimum of the identification functional was proved. Secondly, a theorem on the convergence of the points of the global minimum of the identification functional to the desired true value with an increase in the sample size of observations is taken into account. Assumption of a zero first and restricted second moments of stochastic disturbances in the residual of the equation and observation of variables was made. The convergence of inverse iterations is shown numerically in a model example with significant values of disturbances. The result of the article is new theorems on the conditions of global convergence of computational algorithms with inverse iterations in the problem with mixed disturbances and the justification of possibility of using these algorithms to identify the parameters for discrete stochastic systems of two classes with a simple correspondence between matrix elements and parameters of equivalent systems without state variables