Harnessing graphics processors for the fast computation of acoustic likelihoods in speech recognition

In large vocabulary continuous speech recognition (LVCSR) the acoustic model computations often account for the largest processing overhead. Our weighted finite state transducer (WFST) based decoding engine can utilize a commodity graphics processing unit (GPU) to perform the acoustic computations to move this burden off the main processor. In this paper we describe our new GPU scheme that can achieve a very substantial improvement in recognition speed whilst incurring no reduction in recognition accuracy. We evaluate the GPU technique on a large vocabulary spontaneous speech recognition task using a set of acoustic models with varying complexity and the results consistently show by using the GPU it is possible to reduce the recognition time with largest improvements occurring in systems with large numbers of Gaussians. For the systems which achieve the best accuracy we obtained between 2.5 and 3 times speed-ups. The faster decoding times translate to reductions in space, power and hardware costs by only requiring standard hardware that is already widely installed.     

Computer graphics and medicine: a complex partnership

Developers of graphics applications in medicine face unique challenges in working closely with physicians. They must understand both the technology they support and the limitations of the techniques they use. In this article, I describe the unique challenges facing application developers and review the significant techniques in terms of their strengths and weaknesses in particular applications. In some cases, I also describe lessons learned about misapplying techniques. Hopefully these insights will help to redirect application “misses” and focus effort where it is most needed     

Initial graphics exchange specification: Successes and evolution

The Initial Graphics Exchange Specification (IGES) arrived on the CAD/CAM stage in late 1979 and has enjoyed substantial attention from a broad set of users and implementors since. As the term initial reflected, the first release was only a beginning and IGES continues to be an active organization, as well as an evolving CAD/CAM specification. This paper reviews some of the key events and efforts of the IGES community in furthering the quality and scope of this important interface standard.     

Marine geophysics: Database management and supportive graphics

The National Geophysical Data Center (NGDC), a national and international repository for marine geophysical data, has developed a unique database-management system known as GEODAS (Geophysical Data System). GEODAS is a user-friendly computer system that integrates an index sequential database with an interactive (random access) inventory and data-documentation file. The system is supported by several graphic peripherals, including cathode-ray tube (CRT) terminals, computer-output-microfilm (COM), and plotters. GEODAS incorporates computer software routines that (a) process underway marine geophysical data (i.e. bathymetry, magnetics, and gravity) into a common-format database, (b) provide both graphical and textual inventory information, and (c) retrieve data using multiparameter select criteria. The GEODAS inventory now is used as a remote access system by research institutions throughout the United States. In the future, we hope to provide remote access to the data file as well and to expand the inventory portion to act as an international referral system.     

The render button: How a high-end graphics company developed a personal graphics product

The Personal Visualizer™ was developed to enable a casual user to create high-quality rendered images. Providing sophisticated rendering techniques to casual users forced us to confront a number of issues and to undergo a shift in values. These issues are outlined, and the ways we dealt with them are discussed. Things we would do differently as a result of our development experiences are also discussed.     

Tutorial: time-multiplexed stereoscopic computer graphics

Time-multiplexed stereoscopic systems, which present the stereoscopic image by alternating right- and left-eye views of a scene on a CRT, are considered. The current technology for stereoscopic computer graphics is introduced, and an overview of factors important to the proper construction and display of computer images in a time-multiplexed environment is provided. The discussion covers the computation of stereoscopic views, stereoscopic display algorithms, horizontal parallax and factors affecting image quality     

Computer graphics and displays

A review is given of recent developments in computer graphics, with emphasis on displays. The principles of refresh, storage and raster-scan technologies are outlined, followed by an examination of developments in intelligent raster graphics terminals, displays for 3D computer graphics, and the emergence of powerful personal graphics workstations. Colour hardcopy is needed to complement today's colour raster displays, and ink-jet and electrostatic devices are described. A summary is given of the current status of international standards for computer graphics, and the paper concludes by pointing out some likely areas for future development.     

Telidon graphics and applications

Telidon is essentially a computer graphics code which is very effective as the presentation level protocol for videotex, for which it was originally developed, teletext and other consumer- and business-oriented information systems. Its use of geometric primitives rather than the mosaic characters of its predecessors permits efficient digital storage and rapid transmission of detailed graphics. Display terminals can be built to display the images with different degrees of detail to suit the application.     

Computer graphics visions and challenges: a European perspective

We have briefly described important visions and challenges in computer graphics. They are a personal and therefore subjective selection. But most of these issues have to be addressed and solved - no matter if we call them visions or challenges or something else - if we want to make and further develop computer graphics into a key enabling technology for our IT-based society.     

One-dimensional collision: An interactive graphics program

An interactive program with output to a graphical terminal has been developed for use in an undergraduate physics laboratory. The program computes and draws “snapshots” of one dimensional elastic collisions from data chosen by the student. The aim of the program is to come to a better understanding of the physics of collisions as seen from laboratory and center of mass reference frames. Great stress is laid upon the properties of the latter.     

SPACEBAR: kinematic design by computer graphics

The development of kinematic mechanisms on the design board is often an extremely complex and time-consuming task. The SPACEBAR pre-gram allows the direct design and analysis of mechanisms on the terminal screen. All variables including linkage geometry, stiffness, and applied loading conditions can be input and/or changed from the terminal. The data can be described and displayed in three dimensions. All mechanism configurations can be cycled through their range of travel and viewed in their various geometric positions, again in three dimensions. Output data includes geometric positioning in orthogonal co-ordinates of each node point in the mechanism, velocity and acceleration of the mechanism node points throughout its range of travel, and internal loads and displacements of the individual node points and linkages. All analysis calculations are performed at the v.d.u. and take, at most, a few seconds to complete. Output data can be viewed at the scope and also printed at the discretion of the user.     

G-plot: graphics in the geological computer

G-PLOT is a package of programs for graphic display of data. It has been designed as an integral part of the G-EXEC data-handling system, and adopts the principles of G-EXEC: it is generalized with respect to data, it is structured and modular, and it has a simple (near-English) user interface. In addition, a number of standards have been defined to allow the greatest possible flexibility of usage of the programs, as well as to provide a consistent and powerful set of modules for the applications programmer. The two most significant of these standards are the use of the ‘plot-frame’ concept and the definition of logical raster levels between the data and a ‘logical plotter’. By the use of these standards, it is possible not only to send generated plots to any available graphic device, but also to edit plots by selection and manipulation of logical ‘subframes’—units of plot data such as axes, titling, map frames, and grids. By the combination of G-PLOT facilities with planned enhancements to the G-EXEC system, it will be possible to generate sequences of plot frames from iterative simulation models, or from successively changing parameters (scaling or rotation for example) in normal application programs.     

O-buffer: a framework for sample-based graphics

We present an innovative modeling and rendering primitive, called the O-buffer, as a framework for sample-based graphics. The 2D or 3D O-buffer is, in essence, a conventional image or a volume, respectively, except that samples are not restricted to a regular grid. A sample position in the O-buffer is recorded as an offset to the nearest grid point of a regular base grid (hence the name O-buffer). The O-buffer can greatly improve the expressive power of images and volumes. Image quality can be improved by storing more spatial information with samples and by avoiding multiple resamplings. It can be exploited to represent and render unstructured primitives, such as points, particles, and curvilinear or irregular volumes. The O-buffer is therefore a unified representation for a variety of graphics primitives and supports mixing them in the same scene. It is a semiregular structure which lends itself to efficient construction and rendering. O-buffers may assume a variety of forms including 2D O-buffers, 3D O-buffers, uniform O-buffers, nonuniform O-buffers, adaptive O-buffers, layered-depth O-buffers, and O-buffer trees. We demonstrate the effectiveness of the O--buffer in a variety of applications, such as image-based rendering, point sample rendering, and volume rendering.     

A scalable high-performance graphics processor: GVIP

The GVIP (geometric and TV image processor) graphics processor, which creates and synthesizes computer graphics and TV images and meets the requirements of multi-media systems, is described. The hardware modules that make up this graphics processor include: a 32-bit embedded RISC processor, a Phong and Gouraud shading processor, a texture mapping processor, a hidden surface removal processor, an HDTV video image processor, a BitBlt processor, an imageprocessing module, and an outline font fill generator. These hardware modules fabricated using 0.8 m CMOS standard cells have been placed in three integrated circuit chips. The total number of gates used for one set of chips is approximately 350000.     

Raster graphics display technology: a review

Performance and technologies of monochrome and polychromatic raster CRT visual display units are reviewed. For computer graphics, image quality is affected most by parameters such as size, resolution, luminance, contrast ratio, geometric distortion, flicker and video bandwidth. For polychromatic displays, chromaticity and convergence are also important parameters. Comparisons of 480 mm VDUs show that monochrome untis are less expensive and have better performance except for the use of colour to discriminate between different entities on the screen. Some typical graphics processor output head architectures are reviewed together with trends in display technology.     

RenderMan: pursuing the future of graphics

RenderMan, a powerful, consistent, yet simple 3-D interface that can adequately describe the most common rendering techniques and provides expandability for future growth is described. The RenderMan interface partitions the process of generating realistic images into two distinct areas, modeling and rendering. RenderMan's features are briefly outlined, and the RenderMan Shading language and interface bytestream protocol are examined. Hardware independence and the acceptance of RenderMan are discussed