Publications
Integration of speaker recognition into conversational spoken dialogue systems
September 1, 2003
Conference Paper
Published in:
EUROSPEECH 2003, 1-4 September 2003.
Topic:
R&D area:
Summary
In this paper we examine the integration of speaker identification/verification technology into two dialogue systems developed at MIT: the Mercury air travel reservation system and the Orion task delegation system. These systems both utilize information collected from registered users that is useful in personalizing the system to specific users and that must be securely protected from imposters. Two speaker recognition systems, the MIT Lincoln Laboratory text independent GMM based system and the MIT Laboratory for Computer Science text-constrained speaker-adaptive ASR-based system, are evaluated and compared within the context of these conversational systems.
Summary
In this paper we examine the integration of speaker identification/verification technology into two dialogue systems developed at MIT: the Mercury air travel reservation system and the Orion task delegation system. These systems both utilize information collected from registered users that is useful in personalizing the system to specific users and...
READ MORE
Model compression for GMM based speaker recognition systems
September 1, 2003
Conference Paper
Published in:
EUROSPEECH 2003, 1-4 September 2003.
Topic:
R&D area:
Summary
For large-scale deployments of speaker verification systems models size can be an important issue for not only minimizing storage requirements but also reducing transfer time of models over networks. Model size is also critical for deployments to small, portable devices. In this paper we present a new model compression technique for Gaussian Mixture Model (GMM) based speaker recognition systems. For GMM systems using adaptation from a background model, the compression technique exploits the fact that speaker models are adapted from a single speaker-independent model and not all parameters need to be stored. We present results on the 2002 NIST speaker recognition evaluation cellular telephone corpus and show that the compression technique provides a good tradeoff of compression ratio to performance loss. We are able to achieve a 56:1 compression (624KB -> 11KB) with only a 3.2% relative increase in EER (9.1% -> 9.4%).
Summary
For large-scale deployments of speaker verification systems models size can be an important issue for not only minimizing storage requirements but also reducing transfer time of models over networks. Model size is also critical for deployments to small, portable devices. In this paper we present a new model compression technique...
READ MORE
Measuring the readability of automatic speech-to-text transcripts
September 1, 2003
Conference Paper
Published in:
EUROSPEECH 2003, 1-4 September 2003.
Topic:
R&D area:
Summary
This paper reports initial results from a novel psycholinguistic study that measures the readability of several types of speech transcripts. We define a four-part figure of merit to measure readability: accuracy of answers to comprehension questions, reaction-time for passage reading, reaction-time for question answering and a subjective rating of passage difficulty. We present results from an experiment with 28 test subjects reading transcripts in four experimental conditions.
Summary
This paper reports initial results from a novel psycholinguistic study that measures the readability of several types of speech transcripts. We define a four-part figure of merit to measure readability: accuracy of answers to comprehension questions, reaction-time for passage reading, reaction-time for question answering and a subjective rating of passage...
READ MORE
An examination of wind shear alert integration at the Dallas/Ft. Worth International Airport (DFW)
August 20, 2003
Project Report
Published in:
MIT Lincoln Laboratory Report ATC-309
Summary
The Dallas / Fort Worth International Airport (DFW) is one of the four demonstration system sites for the Integrated Terminal Weather System (ITWS). One of the primary benefits of the ITWS is a suite of algorithms that utilize data from the Terminal Doppler Weather Radar (TDWR) to generate wind shear alerts. DFW also benefits from a Network Expansion of the Low-Level Wind Shear Advisory System (LLWAS-NE). The LLWAS-NE generated alerts are integrated with the radar-based alerts in ITWS to provide Air Traffic Control (ATC) with a comprehensive set of alert information. This study examines the integrated DFW wind shear alerts with emphasis on circumstances in which the detection performance of the TDWR-based wind shear algorithms was poor. Specific detection problems occur in the following situations: when wind shear events over the airport are aligned along a radial to the TDWR, during "non-traditional" wind shear events, when severe signal attenuation occurs during heavy precipitation over the TDWR radar site, and because of excessive TDWR clutter-residue editing over the airport. In all of the cases examined, the LLWAS-NE issued alerts to ATC that would have otherwise gone unreported.
Summary
The Dallas / Fort Worth International Airport (DFW) is one of the four demonstration system sites for the Integrated Terminal Weather System (ITWS). One of the primary benefits of the ITWS is a suite of algorithms that utilize data from the Terminal Doppler Weather Radar (TDWR) to generate wind shear...
READ MORE
Range-velocity ambiguity mitigation schemes for the enhanced Terminal Doppler Weather Radar
August 6, 2003
Conference Paper
Published in:
37th Int. Conf. on Radar Meteorology, 6-12 August 2003.
Summary
The Terminal Doppler Weather Radar (TDWR) radar data acquisition (RDA) subsystem is being replaced as part of a broader FAA program to improve the supportability of the system. An engineering prototype RDA is under development that will provide a modern, open-systems hardware platform and standards-compliant software. The new platform also provides an opportunity to insert algorithms to improve the quality of existing base data products, as well as support future enhancements to the aviation weather services provided by TDWR. There are several outstanding data quality issues with the TDWR. In this paper, we focus on mitigation schemes for the range-velocity ambiguity problem that is especially severe for C-band weather radars such as the TDWR.
Summary
The Terminal Doppler Weather Radar (TDWR) radar data acquisition (RDA) subsystem is being replaced as part of a broader FAA program to improve the supportability of the system. An engineering prototype RDA is under development that will provide a modern, open-systems hardware platform and standards-compliant software. The new platform also...
READ MORE
High-fill-factor, burst-frame-rate charge-coupled device
August 3, 2003
Conference Paper
Published in:
SPIE Vol. 5210, Ultrahigh- and High-Speed Photography, Photonics, and Videography, 3-8 August 2003, pp. 95-104.
Summary
A 512x512-element, multi-frame charge-coupled device (CCD) has been developed for collecting four sequential image frames at megahertz rates. To operate at fast frame rates with high sensitivity, the imager uses an electronic shutter technology developed for back-illuminated CCDs. Device-level simulations were done to estimate the CCD collection well spaces for sub-microsecond photoelectron collection times. Also required for the high frame rates were process enhancements that included metal strapping of the polysilicon gate electrodes and a second metal layer. Tests on finished back-illuminated CCD imagers have demonstrated sequential multi-frame capture capability with integration intervals in the hundreds of nanoseconds range.
Summary
A 512x512-element, multi-frame charge-coupled device (CCD) has been developed for collecting four sequential image frames at megahertz rates. To operate at fast frame rates with high sensitivity, the imager uses an electronic shutter technology developed for back-illuminated CCDs. Device-level simulations were done to estimate the CCD collection well spaces for...
READ MORE
Summary of the EO-1 ALI performance during the first 2.5 years on-orbit
August 3, 2003
Conference Paper
Published in:
SPIE Vol. 5151, Earth Observing Systems VIII, 3-8 August 2003, pp. 574-585.
Topic:
R&D area:
Summary
The Advanced Land Imager (ALI) is a VNIR/SWIR, pushbroom instrument that is flying aboard the Earth Observing-1 (EO-1) spacecraft. Launched on November 21, 2000, the objective of the ALI is to flight validate emerging technologies that can be infused into future land imaging sensors. During the first two and one-half years on-orbit, the performance of the ALI has been evaluated using on-board calibrators and vicarious observations. The results of this evaluation are presented here. The spatial performance of the instrument, derived using stellar, lunar, and bridge observations, is summarized. The radiometric stability of the focal plane and telescope, established using solar, lunar, ground truth, and on-board sources, is also provided.
Summary
The Advanced Land Imager (ALI) is a VNIR/SWIR, pushbroom instrument that is flying aboard the Earth Observing-1 (EO-1) spacecraft. Launched on November 21, 2000, the objective of the ALI is to flight validate emerging technologies that can be infused into future land imaging sensors. During the first two and one-half...
READ MORE
Reducing severe weather delays in congested airspace with weather decision support for tactical air traffic management
June 23, 2003
Conference Paper
Published in:
5th Eurocontrol/DAA ATM R&D Seminar, 23-27 June 2003.
Topic:
R&D area:
R&D group:
Summary
Reducing congested airspace delays due to thunderstorms has become a major objective of the FAA due to the recent growth in convective delays. In 2000 and 2001 the key new initiative for reducing these convective weather delays was "strategic" traffic flow management (TFM) at time scales between 2 and 6 hours in advance using collaborative weather forecasts and routing strategy development. This "strategic" approach experienced difficulties in a large fraction of the weather events because it was not possible to forecast convective storm impacts on routes and capacities accurately enough to accomplish effective traffic flow management. Hence, we proposed in 2001 that there needed to be much greater emphasis on tactical air traffic management at time scales where it would be possible to generate much more accurate convective weather forecasts. In this paper, we describe initial operational results in the very highly congested Great Lakes and Northeast Corridors using weather products from the ongoing Corridor Integrated Weather System (CIWS) concept exploration. Key new capabilities provided by this system include very high update rates (to support tactical air traffic control), much improved echo-tops information, and fully automatic 2-hour convective forecasts using the latest "scale separation" storm tracking technologies. Displays were provided at major terminal areas, en route centers in the corridors, and the FAA Command Center. Substantial reduction in delays has been achieved mostly through weather product usage at the shorter time scales. Quantifying the achieved benefits for this class of products have raised major questions about the conceptual framework for traffic flow management in these congested corridors that must be addressed in the development of air traffic management systems to utilize the weather products.
Summary
Reducing congested airspace delays due to thunderstorms has become a major objective of the FAA due to the recent growth in convective delays. In 2000 and 2001 the key new initiative for reducing these convective weather delays was "strategic" traffic flow management (TFM) at time scales between 2 and 6...
READ MORE
Route selection decision support in convective weather: a case study of the effects of weather and operational assumptions on departure throughput
June 23, 2003
Conference Paper
Published in:
5th Eurocontrol/FAA ATM R&D Seminar, 23-27 2003.
Topic:
R&D area:
R&D group:
Summary
This paper presents a detailed study of a convective weather event affecting the northeastern United States on 19 April 2002: its impacts on departure throughput, the response of traffic managers and an analysis of the potential effects of decision support on system performance. We compare actual departure throughput to what may have been achieved using the Route Availability Planning Tool (RAPT), a prototype decision support tool. We examine two questions: Can decision support identify opportunities to release departures that were missed during the event? How is route selection guidance affected by the operational model incorporated into the decision support tool? By "operational model", we mean three things: the choice of weather forecast information used to define hazards (precipitation, echo tops, etc.), the model for how airspace is used (route definition and allocation) and the assessment of the likelihood that a given route is passable. We focus our analysis on the operational model only; we eliminate weather forecast uncertainty as a factor in the analysis by running RAPT using the actual observed weather as the forecast ('perfect' forecast). Results show that decision support based on perfect forecasts is sensitive to all three elements of the operational model. The sensitivity to weather metrics became evident when we compared decision support based upon perfect forecasts of level 3 vertically integrated liquid (VIL) to that based upon VIL plus storm echo tops. Traffic managers were at times able to move more aircraft by abandoning nominal routing than if they had used nominal routing with perfect weather information. The assessment of route availability will, at times, be ambiguous; different interpretations of that assessment lead to decisions that result in significant differences in departure throughput. These results suggest that for traffic flow management tools, a realistic operational model may be at least as important as the frequently discussed problem of weather forecast uncertainty.
Summary
This paper presents a detailed study of a convective weather event affecting the northeastern United States on 19 April 2002: its impacts on departure throughput, the response of traffic managers and an analysis of the potential effects of decision support on system performance. We compare actual departure throughput to what...
READ MORE
The effect of topography on the initial condition sensitivity of a mesoscale model
June 23, 2003
Conference Paper
Published in:
10th Conf. on Mesoscale Processes, 23-27 June 2003.
Summary
Errors in NWP model forecasts are typically due to deficiencies in the model formulation, inaccuracies associated with the numerical integration techniques, and errors in the specification of initial conditions. This study investigates the latter of these three issues and, in particular, elucidates the errors in the initial conditions due to inadequate data resolution. In a basic sense, for the atmosphere to be adequately sampled at a given length scale, it is not always necessary to increase the number of samples throughout the entire domain. Increased sampling resolution has the greatest benefit in the regions where gradients in the atmospheric conditions exist. Targeted observation techniques attempt to take advantage of this fact by using additional observations to improve the initial analysis in the regions that will have the most impact on forecast accuracy (Emanuel et al. 1995). The result is an economical means to reduce initial condition error and improve forecast accuracy. It is well known that terrain can serve as a localized forcing mechanism in high-resolution models. In addition to acting as a forcing mechanism, variations in terrain can also create strong gradients in the atmospheric fields of models using terrain following vertical coordinates. It is reasonable to assume that if these gradients were better represented in the initial conditions, forecasts accuracies could improve. The present study examines the relationship between terrain variability and the sensitivity of a high-resolution wind forecast to errors in the initial conditions in these areas. The background behind this study and a brief description of the terrain and atmospheric characteristics of the cases used in the experiments are presented in section 2. Initial condition sensitivity analysis results from the fifth generation Pennsylvania State University (PSU), National Center for Atmospheric Research (NCAR) Mesoscale Model (MM5) adjoint and forward models are contained in sections 3 and 4. A summary of the results and conclusions are found in section 5.
Summary
Errors in NWP model forecasts are typically due to deficiencies in the model formulation, inaccuracies associated with the numerical integration techniques, and errors in the specification of initial conditions. This study investigates the latter of these three issues and, in particular, elucidates the errors in the initial conditions due to...
READ MORE