Visual Speech Synthesis by Morphing Visemes

We present MikeTalk, a text-to-audiovisual speech synthesizer which converts input text into an audiovisual speech stream. MikeTalk is built using visemes, which are a small set of images spanning a large range of mouth shapes. The visemes are acquired from a recorded visual corpus of a human subject which is specifically designed to elicit one instantiation of each viseme. Using optical flow methods, correspondence from every viseme to every other viseme is computed automatically. By morphing along this correspondence, a smooth transition between viseme images may be generated. A complete visual utterance is constructed by concatenating viseme transitions. Finally, phoneme and timing information extracted from a text-to-speech synthesizer is exploited to determine which viseme transitions to use, and the rate at which the morphing process should occur. In this manner, we are able to synchronize the visual speech stream with the audio speech stream, and hence give the impression of a photorealistic talking face.     

Integrated circuit implementation of multi-dimensional piecewise-linear functions

In this paper we present an integrated circuit implementing piecewise-linear (PWL) functions with three inputs, where each input can be either analog or digital. The PWL function to be implemented can be chosen by properly storing a set of coefficients in a 4 kB external memory. Experimental results are shown that demonstrate the circuit working up to 50 MHz with a maximum power consumption of 3.7 mW. Measurements corresponding to both static and time-varying inputs are provided and discussed.     

Electrochemically Stimulated Release from Liposomes Embedded in a Polyelectrolyte Multilayer

Triggered release of an entrapped dye from vesicles embedded in a polyelectrolyte multilayer (PEM), as a consequence of the electrochemically induced local pH change in the vicinity of the electrode, is reported. The PEM was deposited on an indium tin oxide (ITO) electrode wherein lipid vesicles filled with a fluorescent dye were embedded. The use of vesicles with a strong negative charge and the polyelectrolyte species of the PEM matrix with a polycation as topmost layer enabled the generation of a stable layer of liposomes in the PEM.     

Additive Manufacturing of Metal Structures at the Micrometer Scale

Currently, the focus of additive manufacturing (AM) is shifting from simple prototyping to actual production. One driving factor of this process is the ability of AM to build geometries that are not accessible by subtractive fabrication techniques. While these techniques often call for a geometry that is easiest to manufacture, AM enables the geometry required for best performance to be built by freeing the design process from restrictions imposed by traditional machining. At the micrometer scale, the design limitations of standard fabrication techniques are even more severe. Microscale AM thus holds great potential, as confirmed by the rapid success of commercial micro‐stereolithography tools as an enabling technology for a broad range of scientific applications. For metals, however, there is still no established AM solution at small scales. To tackle the limited resolution of standard metal AM methods (a few tens of micrometers at best), various new techniques aimed at the micrometer scale and below are presently under development. Here, we review these recent efforts. Specifically, we feature the techniques of direct ink writing, electrohydrodynamic printing, laser‐assisted electrophoretic deposition, laser‐induced forward transfer, local electroplating methods, laser‐induced photoreduction and focused electron or ion beam induced deposition. Although these methods have proven to facilitate the AM of metals with feature sizes in the range of 0.1–10 µm, they are still in a prototype stage and their potential is not fully explored yet. For instance, comprehensive studies of material availability and material properties are often lacking, yet compulsory for actual applications. We address these items while critically discussing and comparing the potential of current microscale metal AM techniques.     

A multiscale model of atherosclerotic plaque formation at its early stage

A multiscale model of atherosclerotic plaque formation at its early stage has been developed in order to integrate the various phenomena leading to fatty streak formation. The different scales considered in this model are in both the spatial domain (from cellular to organism level) and the time domain (from seconds to months). The cellular level was considered by modeling the transport and chemical interactions of low-density lipoproteins (LDL) and other agents in a stenosed artery. This was linked to arterial thickening (organ level). Mean blood LDL level (organism level) was selected as one of the critical factors for atherosclerotic formation along with wall shear stress (WSS) exerted on the endothelium (the inner portion of the artery). It was observed that plaque location was dependent on WSS and that plaque size, number, and growth was dependent on mean blood LDL levels as well.     

Introduction: Learning and Vision at CBCL

This special issue is dedicated to some of the recent work at the Center for Biological and Computational Learning at MIT in applying machine learning techniques to computer vision.     

Cooperative Vaccinia Infection Demonstrated at the Single-Cell Level Using FluidFM

The mechanisms used by viruses to enter and replicate within host cells are subjects of intense investigation. These studies are ultimately aimed at development of new drugs that interfere with these processes. Virus entry and infection are generally monitored by dispensing bulk virus suspensions on layers of cells without accounting for the fate of each virion. Here, we take advantage of the recently developed FluidFM to deposit single vaccinia virions onto individual cells in a controlled manner. While the majority of virions were blocked prior to early gene expression, infection of individual cells increased in a nondeterministic fashion with respect to the number of viruses placed. Microscopic analyses of several stages of the virus lifecycle indicated that this was the result of cooperativity between virions during early stages of infection. These findings highlight the importance of performing controlled virus infection experiments at the single cell level.     

A Trainable System for Object Detection

This paper presents a general, trainable system for object detection in unconstrained, cluttered scenes. The system derives much of its power from a representation that describes an object class in terms of an overcomplete dictionary of local, oriented, multiscale intensity differences between adjacent regions, efficiently computable as a Haar wavelet transform. This example-based learning approach implicitly derives a model of an object class by training a support vector machine classifier using a large set of positive and negative examples. We present results on face, people, and car detection tasks using the same architecture. In addition, we quantify how the representation affects detection performance by considering several alternate representations including pixels and principal components. We also describe a real-time application of our person detection system as part of a driver assistance system.