Personalising speech-to-speech translation: Unsupervised cross-lingual speaker adaptation for HMM-based speech synthesis

In this paper we present results of unsupervised cross-lingual speaker adaptation applied to text-to-speech synthesis. The application of our research is the personalisation of speech-to-speech translation in which we employ a HMM statistical framework for both speech recognition and synthesis. This framework provides a logical mechanism to adapt synthesised speech output to the voice of the user by way of speech recognition. In this work we present results of several different unsupervised and cross-lingual adaptation approaches as well as an end-to-end speaker adaptive speech-to-speech translation system. Our experiments show that we can successfully apply speaker adaptation in both unsupervised and cross-lingual scenarios and our proposed algorithms seem to generalise well for several language pairs. We also discuss important future directions including the need for better evaluation metrics.     

Lattice Constants and Compositions of Manganese-Bearing Ferrites

When manganese-bearing ferrites (Mn _x Fe_{3- x} O_4-delta, with x = 0.17-1.09) were prepared in the present study via a solid-state reaction, manganese ions were incorporated into the lattice sites of the ferrites in the divalent state. The lattice constant ( a ₀) of those ferrites increased linearly as the manganese composition x increased. A ferrite with x = 1.28 contained manganese ions with a valence state higher than 3+; however, no further increase in a ₀ occurred at valences higher than 2+.     

Post-irradiation studies on knock-out and pseudo-recoil releases of fission products from fissioning UO2

By using post-irradiation techniques, in-pile releases of ¹³³Xe, ^85mKr, ⁸⁸Kr, ⁸⁷Kr and ¹³⁸Xe from UO₂ fissioning at low temperatures below about 200° C are studied: these are analyzed into a time-dependent knock-out and time-independent pseudo-recoil releases. For the latter, a “self knock-out” mechanism is proposed: when a fission fragment loses thoroughly its energy near the UO₂ surface and stops there, it will knock out the surface substances and accordingly the fragment (i.e. the fission product) will be released. The effective thickness of the layer where the self knock-out occurs is found to be ~7Å. As for the knock-out release, the following is estimated from its dependence on various factors: the knock-out release of fission products occurs from the surface layer with the effective thickness of ~20Å: the shape of UO₂ matrix knocked out by one fission fragment passing through the surface is equivalent to a cylinder ~32Å diameter by ~27Å thick, (i.e. the knock-out coefficient for UO₂ is ~660 uranium atoms per knock-out event). On the basis of the above estimations, the conclusions derived from the past in-pile studies of fission gas releases are evaluated.     

Order-disorder transition and its effect on microwave quality factor Q in Ba(Zn1/3Nb2/3)O3 system

Influences of composition deviation from stoichiometry and heat treatment on crystal phases and Q factor in Ba(Zn_1/3Nb_2/3)O₃(BZN) were studied. The structural order and the crystal phases strongly depended on the slight composition deviation from stoichiometric BZN. The maximum Q factor was obtained at the vicinity of the stoichiometric BZN. In the other regions, non-stoichiometric disordered BZN or ordered BZN with secondary phase were formed, and their Q factors were found to be low. For the stoichiometric BZN, the order-disorder phase transition occurred between 1300 and 1400^°C. The crystal-structural ordering of the stoichiometric BZN was improved by post-annealing at below its transition temperature, conserving the density and the grain size. However, no significant Q factor improvement was found. The Q factor of the stoichiometric BZN strongly depends on the density and grain size not on the crystal-structural order. These results suggest that the ceramic microstructure such as the pore and grain boundary, the secondary phase and lattice defect caused by non-stoichiometry affect the variation of the Q factor in BZN system than the crystal-structural ordering.     

Insights from ab initio molecular dynamics simulations for a multicomponent oxide glass

It remains a challenge to establish structural models of multicomponent oxide glass systems. In this study, we have investigated 68.3SiO₂–16.1B₂O₃–4.2Al₂O₃–11.4Na₂O glass and melt structures by ab initio molecular dynamics (AIMD) simulations. The atomic configurations obtained from AIMD simulations were validated against ¹⁷O solid‐state NMR spectrum under 24.0 T and neutron diffraction data, and excellent agreement was achieved. The bond lengths, angles, and coordination geometries were statistically analyzed for each atomic species. Here we particularly address the role of minor atomic species such as five‐coordinate Si (Si^V) and Al (Al^V). The Si^V–O bond lengths and O–Si^V–O angle distribution in the glass indicated 1.718 Å and three peaks at 90°, 120°, and 175°, which are assigned to a coordination geometry of the trigonal bipyramidal structure. Ring statistic analysis revealed that Si^V and Al^V were found to preferentially contribute to the formation of small ring sizes.     

Initial impression versus missing information as explanations of the set-size effect.

The averaging model of information integration generally predicts a decelerating set-size effect (SSE), because it assumes that an "initial neutral impression" is always averaged with the information items that are presented. However, a modified averaging model, called the path-analytic averaging (PAA) model, predicts that the SSE will not always occur. What is considered the "initial impression" in the averaging model is reconceptualized as "inferred missing information" in the PAA model. When the number of presented items equals the number of items deemed important and both increase together, the PAA model predicts that there will be no SSE because there is no missing information. When the number of presented items increases, so that the added items provide information previously missing, a SSE should occur. These predictions of the PAA model were tested in an experiment in which 36 undergraduates rated the desirability of candidates for secretarial positions based on 1, 2, or 3 items of information. For most Ss, the PAA predictions were confirmed; for some Ss, however, the results are inconsistent with both the PAA model and the usual averaging model predictions. The latter Ss were distinguished from the others by their apparent use of scores less than the scale midpoint when they inferred information that was considered important but missing. (6 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Optimal impedance control based on body inertia for a hydraulically driven hexapod robot walking on uneven and extremely soft terrain

This paper presents the implementation of impedance control for a hydraulically driven hexapod robot named COMET‐IV, which can walk on uneven and extremely soft terrain. To achieve the dynamic behavior of the hexapod robot, changes in center of mass and body attitude must be taken into consideration during the walking periods. Indirect force control via impedance control is used to address these issues. Two different impedance control schemes are developed and implemented: single‐leg impedance control and center of mass‐‐based impedance control. In the case of single‐leg impedance control, we derive the necessary impedance and adjust parameters (mass, damping, and stiffness) according to the robot legs' configuration. For center of mass–based impedance control, we use the sum of the forces of the support legs as a control input (represented by the body's current center of mass) for the derived impedance control and adjust parameters based on the robot body's configuration. The virtual forces from the robot body's moment of inertia are adapted to achieve optimal control via a linear quadratic regulator method for the proposed indirect attitude control. In addition, a compliant switching mechanism is designed to ensure that the implementation of the controller is applicable to the tripod sequences of force‐based walking modules. Evaluation and verification tests were conducted in the laboratory and the actual field with uneven terrain and extremely soft surfaces. © 2011 Wiley Periodicals, Inc.     

Guidance and nonlinear control system for autonomous flight of minirotorcraft unmanned aerial vehicles

Small unmanned aerial vehicles (UAVs) are becoming popular among researchers and vital platforms for several autonomous mission systems. In this paper, we present the design and development of a miniature autonomous rotorcraft weighing less than 700 g and capable of waypoint navigation, trajectory tracking, visual navigation, precise hovering, and automatic takeoff and landing. In an effort to make advanced autonomous behaviors available to mini‐ and microrotorcraft, an embedded and inexpensive autopilot was developed. To compensate for the weaknesses of the low‐cost equipment, we put our efforts into designing a reliable model‐based nonlinear controller that uses an inner‐loop outer‐loop control scheme. The developed flight controller considers the system's nonlinearities, guarantees the stability of the closed‐loop system, and results in a practical controller that is easy to implement and to tune. In addition to controller design and stability analysis, the paper provides information about the overall control architecture and the UAV system integration, including guidance laws, navigation algorithms, control system implementation, and autopilot hardware. The guidance, navigation, and control (GN&C) algorithms were implemented on a miniature quadrotor UAV that has undergone an extensive program of flight tests, resulting in various flight behaviors under autonomous control from takeoff to landing. Experimental results that demonstrate the operation of the GN&C algorithms and the capabilities of our autonomous micro air vehicle are presented. © 2009 Wiley Periodicals, Inc.     

Scaling up experiments of honeycomb catalysts for oxidation of hydrogen and methane gases

For a large volume air cleanup system in nuclear fusion facility honeycomb type oxidizing catalyst offers more useful advantage in terms of their low pressure drop than the conventional particle packed bed. Scaling up examination of 35 times large volume Pd catalyst deposit on metal honeycomb Al–Cr-steel alloy had been performed comparing with fundamental small size honeycomb. The catalyst was heated up to over 400 °C with stepwise by combined the inner and external heaters under various flow rate of processing gas containing various concentrations of hydrogen and methane. Then oxidizing reaction rates and the catalytic activation energies were evaluated and thermal property of catalyst bed was estimated considering exothermic oxidizing reaction in the process gas. As a result it was revealed that the scaling up honeycomb catalyst bed would be practically useful for the air cleanup system.     

Analysis of Speaker Adaptation Algorithms for HMM-Based Speech Synthesis and a Constrained SMAPLR Adaptation Algorithm

In this paper, we analyze the effects of several factors and configuration choices encountered during training and model construction when we want to obtain better and more stable adaptation in HMM-based speech synthesis. We then propose a new adaptation algorithm called constrained structural maximum a posteriori linear regression (CSMAPLR) whose derivation is based on the knowledge obtained in this analysis and on the results of comparing several conventional adaptation algorithms. Here, we investigate six major aspects of the speaker adaptation: initial models; the amount of the training data for the initial models; the transform functions, estimation criteria, and sensitivity of several linear regression adaptation algorithms; and combination algorithms. Analyzing the effect of the initial model, we compare speaker-dependent models, gender-independent models, and the simultaneous use of the gender-dependent models to single use of the gender-dependent models. Analyzing the effect of the transform functions, we compare the transform function for only mean vectors with that for mean vectors and covariance matrices. Analyzing the effect of the estimation criteria, we compare the ML criterion with a robust estimation criterion called structural MAP. We evaluate the sensitivity of several thresholds for the piecewise linear regression algorithms and take up methods combining MAP adaptation with the linear regression algorithms. We incorporate these adaptation algorithms into our speech synthesis system and present several subjective and objective evaluation results showing the utility and effectiveness of these algorithms in speaker adaptation for HMM-based speech synthesis.