Style-Controllable Speech-Driven Gesture Synthesis Using Normalising Flows

Automatic synthesis of realistic gestures promises to transform the fields of animation, avatars and communicative agents. In off-line applications, novel tools can alter the role of an animator to that of a director, who provides only high-level input for the desired animation; a learned network then translates these instructions into an appropriate sequence of body poses. In interactive scenarios, systems for generating natural animations on the fly are key to achieving believable and relatable characters. In this paper we address some of the core issues towards these ends. By adapting a deep learning-based motion synthesis method called MoGlow, we propose a new generative model for generating state-of-the-art realistic speech-driven gesticulation. Owing to the probabilistic nature of the approach, our model can produce a battery of different, yet plausible, gestures given the same input speech signal. Just like humans, this gives a rich natural variation of motion. We additionally demonstrate the ability to exert directorial control over the output style, such as gesture level, speed, symmetry and spacial extent. Such control can be leveraged to convey a desired character personality or mood. We achieve all this without any manual annotation of the data. User studies evaluating upper-body gesticulation confirm that the generated motions are natural and well match the input speech. Our method scores above all prior systems and baselines on these measures, and comes close to the ratings of the original recorded motions. We furthermore find that we can accurately control gesticulation styles without unnecessarily compromising perceived naturalness. Finally, we also demonstrate an application of the same method to full-body gesticulation, including the synthesis of stepping motion and stance.     

Uzawa Conjugate Gradient Domain Decomposition Methods for Coupled Stokes Flows

This paper deals with nonoverlapping domain decomposition methods for two coupled Stokes flows, based on the duality theory. By introducing a fictitious variable in the transmission condition and using saddle-point equations, the problem is restated as a linearly constrained maximization problem. According to whether constraints are uncoupled Stokes problems or uncoupled Poisson problems, two Uzawa-type domain decomposition algorithms are proposed. The results of some numerical experiments on a model problem are given.     

Formalization of Shannon’s Theorems

The most fundamental results of information theory are Shannon’s theorems. These theorems express the bounds for (1) reliable data compression and (2) data transmission over a noisy channel. Their proofs are non-trivial but are rarely detailed, even in the introductory literature. This lack of formal foundations is all the more unfortunate that crucial results in computer security rely solely on information theory: this is the so-called “unconditional security”. In this article, we report on the formalization of a library for information theory in the SSReflect extension of the Coq proof-assistant. In particular, we produce the first formal proofs of the source coding theorem, that introduces the entropy as the bound for lossless compression, and of the channel coding theorem, that introduces the capacity as the bound for reliable communication over a noisy channel.     

Frequency-domain identification of continuous-time ARMA models from sampled data

The subject of this paper is the direct identification of continuous-time autoregressive moving average (CARMA) models. The topic is viewed from the frequency domain perspective which then turns the reconstruction of the continuous-time power spectral density (CT-PSD) into a key issue. The first part of the paper therefore concerns the approximate estimation of the CT-PSD from uniformly sampled data under the assumption that the model has a certain relative degree. The approach has its point of origin in the frequency domain Whittle likelihood estimator. The discrete- or continuous-time spectral densities are estimated from equidistant samples of the output. For low sampling rates the discrete-time spectral density is modeled directly by its continuous-time spectral density using the Poisson summation formula. In the case of rapid sampling the continuous-time spectral density is estimated directly by modifying its discrete-time counterpart.     

27Al and 29Si NMR study of sol-gel derived aluminosilicates and sodium aluminosilicates

Solid state ²⁷Al and ²⁹Si NMR was used to examine the structures of aluminosilicates and sodium aluminosilicates prepared by the sol-gel method from metal alkoxides. In contrast to the borosilicate system, where B-O-Si bonds are not formed until heat treatment above 150° C, Al-O-Si formation appears complete upon gelation. Aluminium occupies tetrahedral [AlO₄]^– sites in the polymer network and octahedral [Al(H₂O)₆]³⁺ (or similar) sites in the intersticies for charge balance. When sodium is added as a counter ion the octahedral aluminium is converted to tetrahedral aluminium in the oxide network. In gels of high aluminium content prepared from (Bu^sO)₂Al-O-Si(OEt)₃, some aluminium in five coordinate environments is also observed. All gels remain amorphous on heating to 800° C.     

Nonlinear black-box modeling in system identification: a unified overview

A nonlinear black-box structure for a dynamical system is a model structure that is prepared to describe virtually any nonlinear dynamics. There has been considerable recent interest in this area, with structures based on neural networks, radial basis networks, wavelet networks and hinging hyperplanes, as well as wavelet-transform-based methods and models based on fuzzy sets and fuzzy rules. This paper describes all these approaches in a common framework, from a user's perspective. It focuses on what are the common features in the different approaches, the choices that have to be made and what considerations are relevant for a successful system-identification application of these techniques. It is pointed out that the nonlinear structures can be seen as a concatenation of a mapping form observed data to a regression vector and a nonlinear mapping from the regressor space to the output space. These mappings are discussed separately. The latter mapping is usually formed as a basis function expansion. The basis functions are typically formed from one simple scalar function, which is modified in terms of scale and location. The expansion from the scalar argument to the regressor space is achieved by a radial- or a ridge-type approach. Basic techniques for estimating the parameters in the structures are criterion minimization, as well as two-step procedures, where first the relevant basis functions are determined, using data, and then a linear least-squares step to determine the coordinates of the function approximation. A particular problem is to deal with the large number of potentially necessary parameters. This is handled by making the number of ‘used’ parameters considerably less than the number of ‘offered’ parameters, by regularization, shrinking, pruning or regressor selection.     

Abstraction paradigms for computer graphics

In computer graphics we use techniques from different areas of mathematics. Mathematical models are used to simulate real-world objects, as well as natural phenomena. In order to understand these models and pose relevant problems in each particular field of this area, it is important to create levels of abstraction. These levels encapsulate common properties of the different models and allow us to have a global, conceptual view of the methods and techniques in each field. In this paper we study a paradigm for creating abstraction levels that can also be used to characterize more general problems in computational applied mathematics. We apply this paradigm to different areas of computer graphics: modeling, animation, illumination, color theory, image processing and human-computer interface.     

Creep and rupture behaviour in fibre bundles

The time dependent rupture of a fibre bundle under constant tensile load and with scatter in fibre stiffness and fibre strength is analysed using a general formulation of the creep law. The total rupture time is divided into an incubation time, with all fibres intact, and a failure propagation time where fibres fail progressively. Governing equations are given for both time intervals. It is shown that the failure propagation time interval is independent of the load if no spatial stress redistribution occurs and all fibres remain intact upon loading. For a bundle consisting of originally uniform fibres where a fraction is afflicted with a certain defect, the time intervals are calculated assuming Norton's non linear creep law.     

Numerical simulation of segmented penetrator impact

The performance of segmented and continuous penetrators impacting semi-infinite and spaced armor is studied using both the EPIC-2 and HULL hydrocodes. First the performance of a segmented rod is studied, striking semi-infinite armor, for various parameters such as striking velocity, segment spacing and number of segments. Then an actual penetrator configuration proposed by A. Charters is analyzed and the use of normalized penetration is discussed. Finally three-dimensional simulations are presented for segmented and continuous penetrators impacting oblique spaced armor varying such parameters as striking velocity, segment spacing, number of segments, and target thickness.     

Human fear conditioning is related to dopaminergic and serotonergic biological markers.

Biological markers for acquisition and extinction of fear conditioning were studied in 40 individuals selected for displaying either good or poor acquisiton of fear conditioning, as estimated by the skin conductance response. Participants with a short serotonin transporter (5-HTT) promoter allele or low monoamine oxidase activity in platelets (trbc-MAO) displayed better acquisition than those with only long alleles or high trbc-MAO, whereas participants with a long dopamine D4 receptor (D4DR) exon III allele showed delayed extinction compared with those with only short alleles. The findings, that D4DR exon III and 5-HTT promotor genotypes and trbc-MAO activity are related to human fear conditioning, a basic form of associative learning, are consistent with animal studies suggesting a genetic contribution to fear conditioning. The authors suggest that in humans these genetic mechanisms are partly dopaminergic and serotonergic in origin. (PsycINFO Database Record (c) 2010 APA, all rights reserved)