A Cascade Visual Front End for Speaker Independent Automatic Speechreading

We propose a three-stage pixel-based visual front end for automatic speechreading (lipreading) that results in significantly improved recognition performance of spoken words or phonemes. The proposed algorithm is a cascade of three transforms applied on a three-dimensional video region-of-interest that contains the speaker's mouth area. The first stage is a typical image compression transform that achieves a high-energy, reduced-dimensionality representation of the video data. The second stage is a linear discriminant analysis-based data projection, which is applied on a concatenation of a small amount of consecutive image transformed video data. The third stage is a data rotation by means of a maximum likelihood linear transform that optimizes the likelihood of the observed data under the assumption of their class-conditional multivariate normal distribution with diagonal covariance. We applied the algorithm to visual-only 52-class phonetic and 27-class visemic classification on a 162-subject, 8-hour long, large vocabulary, continuous speech audio-visual database. We demonstrated significant classification accuracy gains by each added stage of the proposed algorithm which, when combined, can achieve up to 27% improvement. Overall, we achieved a 60% (49%) visual-only frame-level visemic classification accuracy with (without) use of test set viseme boundaries. In addition, we report improved audio-visual phonetic classification over the use of a single-stage image transform visual front end. Finally, we discuss preliminary speech recognition results.     

A derivative-free distributed filtering approach for sensorless control of nonlinear systems

This article examines the problem of sensorless control for nonlinear dynamical systems with the use of derivative-free Extended Information Filtering (EIF). The system is first subject to a linearisation transformation and next state estimation is performed by applying the standard Kalman Filter to the linearised model. At a second level, the standard Information Filter is used to fuse the state estimates obtained from local derivative-free Kalman filters running at the local information processing nodes. This approach has significant advantages because unlike the EIF (i) is not based on local linearisation of the nonlinear dynamics (ii) does not assume truncation of higher order Taylor expansion terms thus preserving the accuracy and robustness of the performed estimation and (iii) does not require the computation of Jacobian matrices. As a case study a robotic manipulator is considered and a cameras network consisting of multiple vision nodes is assumed to provide the visual information to be used in the control loop. A derivative-free implementation of the EIF is used to produce the aggregate state vector of the robot by processing local state estimates coming from the distributed vision nodes. The performance of the considered sensorless control scheme is evaluated through simulation experiments.     

Nonlinear Kalman Filters and Particle Filters for integrated navigation of unmanned aerial vehicles

The paper studies and compares nonlinear Kalman Filtering methods and Particle Filtering methods for estimating the state vector of Unmanned Aerial Vehicles (UAVs) through the fusion of sensor measurements. Next, the paper proposes the use of the estimated state vector in a control loop for autonomous navigation and trajectory tracking by the UAVs. The proposed nonlinear controller is derived according to the flatness-based control theory. The estimation of the UAV’s state vector is carried out with the use of (i) Extended Kalman Filtering (EKF), (ii) Sigma-Point Kalman Filtering (SPKF), (iii) Particle Filtering (PF), and (iv) a new nonlinear estimation method which is the Derivative-free nonlinear Kalman Filtering (DKF). The performance of the nonlinear control loop which is based on these nonlinear state estimation methods is evaluated through simulation tests. Comparing the aforementioned filtering methods in terms of estimation accuracy and computation speed, it is shown that the Sigma-Point Kalman Filtering is a reliable and computationally efficient approach to state estimation-based control, while Particle Filtering is well-suited to accommodate non-Gaussian measurements. Moreover, it is shown that the Derivative-free nonlinear Kalman Filter is faster than the rest of the nonlinear filters while also succeeding accurate, in terms of variance, state estimates.     

Charge transport mechanisms and memory effects in amorphous TaNx thin films

Amorphous semiconducting materials have unique electrical properties that may be beneficial in nanoelectronics, such as low leakage current, charge memory effects, and hysteresis functionality. However, electrical characteristics between different or neighboring regions in the same amorphous nanostructure may differ greatly. In this work, the bulk and surface local charge carrier transport properties of a-TaN_x amorphous thin films deposited in two different substrates are investigated by conductive atomic force microscopy. The nitride films are grown either on Au (100) or Si [100] substrates by pulsed laser deposition at 157 nm in nitrogen environment. For the a-TaN_x films deposited on Au, it is found that they display a negligible leakage current until a high bias voltage is reached. On the contrary, a much lower threshold voltage for the leakage current and a lower total resistance is observed for the a-TaN_x film deposited on the Si substrate. Furthermore, I-V characteristics of the a-TaN_x film deposited on Au show significant hysteresis effects for both polarities of bias voltage, while for the film deposited on Si hysteresis, effects appear only for positive bias voltage, suggesting that with the usage of the appropriate substrate, the a-TaN_x nanodomains may have potential use as charge memory devices.     

On the occasional biodegradation of pharmaceuticals in the activated sludge process: the example of the antibiotic sulfamethoxazole

Sulfamethoxazole, a common antibiotic, was found to be biodegradable under aerobic conditions. The fate of sulfamethoxazole in the activated sludge process was studied using a Sequencing Batch Reactor (SBR). Aerobic biomass was acclimated to sulfamethoxazole and a series of kinetic experiments were conducted to investigate the impact of other carbon and nitrogen sources on the degradation of the antibiotic. It was found that sulfamethoxazole serves both as carbon and nitrogen source for the enriched consortium. It was degraded whenever there was a depletion of carbon or nitrogen or both in the feeding medium, while in the presence of acetate and ammonium nitrogen (alternative carbon and nitrogen sources, respectively), sulfamethoxazole remained intact.     

Discrimination and identification of dynamic catalytic reaction models via introduction of feedback

Linear and nonlinear feedback manipulations based on effluent concentration measurements from a catalytic CSTR are successfully used to discriminate between two rival models for catalytic N₂O decomposition. Introduction of feedback causes steady-state multiplicity and bifurcation to periodic solutions. Much greater differences between the alternative models are revealed by this approach than by steady-state and “transient response” methods. Limit cycles obtained with feedback have in some cases nonharmonic form near bifurcation which may be attributed to bifurcation phenomena of higher order than the usual Hopf case. The dynamic studies are systematically organized using a new “decrement diagram” technique which displays the influence of two feedback parameters on closed-loop system bifurcations.