Publications
A study of computation speed-ups of the GMM-UBM speaker recognition system
September 5, 1999
Conference Paper
Published in:
6th European Conf. on Speech Communication and Technology, EUROSPEECH, 5-9 September 1999.
Topic:
R&D area:
Summary
The Gaussian Mixture Model Universal Background Model (GMM-UBM) speaker recognition system has demonstrated very high performance in several NIST evaluations. Such evaluations, however, are concerned only with classification accuracy. In many applications, system effectiveness must be evaluated in light of both accuracy and execution speed. We present here a number of techniques for decreasing computation. Using data from the Switchboard telephone speech corpus, we show that significant speed-ups can be obtained while sacrificing surprisingly little accuracy. We expect that these techniques, involving lowering model order as well as processing fewer speech frames, will apply equally well to other recognition systems.
Summary
The Gaussian Mixture Model Universal Background Model (GMM-UBM) speaker recognition system has demonstrated very high performance in several NIST evaluations. Such evaluations, however, are concerned only with classification accuracy. In many applications, system effectiveness must be evaluated in light of both accuracy and execution speed. We present here a number...
READ MORE
Evaluation of confidence measures for language identification
September 5, 1999
Conference Paper
Published in:
6th European Conf. on Speech Communication and Technology, EUROSPEECH, 5-9 September 1999.
Topic:
R&D area:
Summary
In this paper we examine various ways to derive confidence measures for a language identification system, using phone recognition followed by language models, and describe the application of an evaluation metric for measuring the "goodness" of the different confidence measures. Experiments are conducted on the 1996 NIST Language Identification Evaluation corpus (derived from the Callfriend corpus of conversational telephone speech). The system is trained on the NIST 96 development data and evaluated on the NIST 96 evaluation data. Results indicate that we are able to predict the performance of a system and quantitatively evaluate how well the prediction holds on new data.
Summary
In this paper we examine various ways to derive confidence measures for a language identification system, using phone recognition followed by language models, and describe the application of an evaluation metric for measuring the "goodness" of the different confidence measures. Experiments are conducted on the 1996 NIST Language Identification Evaluation...
READ MORE
Speaker and language recognition using speech codec parameters
September 5, 1999
Conference Paper
Published in:
EUROSPEECH 99, 5-10 September 1999.
Topic:
R&D area:
Summary
In this paper, we investigate the effect of speech coding on speaker and language recognition tasks. Three coders were selected to cover a wide range of quality and bit rates: GSM at 12.2 kb/s, G.729 at 8 kb/s, and G.723.1 at 5.3 kb/s. Our objective is to measure recognition performance from either the synthesized speech or directly from the coder parameters themselves. We show that using speech synthesized from the three codecs, GMM-based speaker verification and phone-based language recognition performance generally degrades with coder bit rate, i.e., from GSM to G.729 to G.723.1, relative to an uncoded baseline. In addition, speaker verification for all codecs shows a performance decrease as the degree of mismatch between training and testing conditions increases, while language recognition exhibited no decrease in performance. We also present initial results in determining the relative importance of codec system components in their direct use for recognition tasks. For the G.729 codec, it is shown that removal of the post- filter in the decoder helps speaker verification performance under the mismatched condition. On the other hand, with use of G.729 LSF-based mel-cepstra, performance decreases under all conditions, indicating the need for a residual contribution to the feature representation.
Summary
In this paper, we investigate the effect of speech coding on speaker and language recognition tasks. Three coders were selected to cover a wide range of quality and bit rates: GSM at 12.2 kb/s, G.729 at 8 kb/s, and G.723.1 at 5.3 kb/s. Our objective is to measure recognition performance...
READ MORE
Modeling of the glottal flow derivative waveform with application to speaker identification
September 1, 1999
Journal Article
Published in:
IEEE Trans. Speech Audio Process., Vol. 7, No. 5, September 1999, pp. 569-586.
Topic:
R&D area:
Summary
An automatic technique for estimating and modeling the glottal flow derivative source waveform from speech, and applying the model parameters to speaker identification, is presented. The estimate of the glottal flow derivative is decomposed into coarse structure, representing the general flow shape, and fine structure, comprising aspiration and other perturbations in the flow, from which model parameters are obtained. The glottal flow derivative is estimated using an inverse filter determined within a time interval of vocal-fold closure that is identified through differences in formant frequency modulation during the open and closed phases of the glottal cycle. This formant motion is predicted by Ananthapadmanabha and Fant to be a result of time-varying and nonlinear source/vocal tract coupling within a glottal cycle. The glottal flow derivative estimate is modeled using the Liljencrants-Fant model to capture its coarse structure, while the fine structure of the flow derivative is represented through energy and perturbation measures. The model parameters are used in a Gaussian mixture model speaker identification (SID) system. Both coarse- and fine-structure glottal features are shown to contain significant speaker-dependent information. For a large TIMIT database subset, averaging over male and female SID scores, the coarse-structure parameters achieve about 60% accuracy, the fine-structure parameters give about 40% accuracy, and their combination yields about 70% correct identification. Finally, in preliminary experiments on the counterpart telephone-degraded NTIMIT database, about a 5% error reduction in SID scores is obtained when source features are combined with traditional mel-cepstral measures.
Summary
An automatic technique for estimating and modeling the glottal flow derivative source waveform from speech, and applying the model parameters to speaker identification, is presented. The estimate of the glottal flow derivative is decomposed into coarse structure, representing the general flow shape, and fine structure, comprising aspiration and other perturbations...
READ MORE
Criteria for sprites and elves based on Schumann resonance observations
July 27, 1999
Journal Article
Published in:
J. Geophys. Res., Vol. 104, No. D14, 27 July 1999, pp. 16,943-16,964.
Summary
Ground flashes with positive polarity associated with both sprites and elves excite the Earth's Schumann resonances to amplitudes several times greater than the background resonances. Theoretical predictions for dielectric breakdown in the mesosphere are tested using ELF methods to evaluate vertical charge moments of positive ground flashes. Comparisons of the measured time constants for lightning charge transfer with the electrostatic relaxation time at altitudes of nighttime sprite initiation (50-70 km) generally validate the electrostatic assumption in predictions made initially by Wilson [1925]. The measured charge moments (Q dS = 200-2000 C-km) are large in comparison with ordinary negative lightning but are generally insufficient to account for conventional air breakdown at sprite altitudes. The measured charge moments, however, are sufficient to account for electron runaway breakdown, and the long avalanche length in this mechanism also accounts for the exclusive association of sprites with ground flashes of positive polarity. The association of elves with large peak currents (50-200 kA) measured by the National Lightning Detection Network in a band pass beyond the Schumann resonance range is consistent with an electromagnetic pulse mechanism for these events.
Summary
Ground flashes with positive polarity associated with both sprites and elves excite the Earth's Schumann resonances to amplitudes several times greater than the background resonances. Theoretical predictions for dielectric breakdown in the mesosphere are tested using ELF methods to evaluate vertical charge moments of positive ground flashes. Comparisons of the...
READ MORE
The FAA Terminal Convective Weather Forecast product: scale separation filter optimization
July 12, 1999
Conference Paper
Published in:
29th Int. Conf. on Radar Meteorology, 12-16 July 1999.
Topic:
R&D area:
R&D group:
Summary
A large percentage of serious air traffic delay at major airports in the warm season is caused by convective weather. The FAA Convective Weather Product Development team (PDT) has developed a Terminal Convective Weather Forecast product (TCWF) that can account for short-term (out to 60 min) systematic growth and decay of thunderstorms. The team began work three years ago by evaluating air traffic user needs and requirements. We found that users were willing to trade off forecast accuracy for longer lead times, especially for air traffic management plans that were easy to implement or that incurred low risk (Forman, et al., 1999). The PDT was able to develop an operationally useful forecast product that has been demonstrated in Dallas, TX since March, 1998 (Hallowell, et al., 1999). Further improvements have been made, and testing is now taking place at both Dallas and Orlando, FL. This paper summarizes the basic algorithm methodology and presents quantitative results on optimization of the scale separation filter, which is an integral aspect of the forecast algorithm.
Summary
A large percentage of serious air traffic delay at major airports in the warm season is caused by convective weather. The FAA Convective Weather Product Development team (PDT) has developed a Terminal Convective Weather Forecast product (TCWF) that can account for short-term (out to 60 min) systematic growth and decay...
READ MORE
Wind shear detection using the Next Generation Airport Surveillance Radar
July 12, 1999
Project Report
Published in:
MIT Lincoln Laboratory Report ATC-266
Summary
The Federal Aviation Administration (FAA) is deploying a Weather Systems Processor (WSP) for the current-generation Airport Surveillance Radar - ASR-9. This modification exploits the coherency of the ASR-9 to perform Doppler wind measurement. Signature recognition algorithms then automatically detect low altitude wind shear phenomena, track thunderstorm motion and display appropriate graphical and alphanumeric alerts to air traffic control (ATC) personnel. The FAA and U.S. Air Force are now procuring an ASR-11 to replace older terminal surveillance radars at facilities that did not receive the ASR-9. Although the antenna pattern, scan rate and energy-on-target of the ASR-11 match the corresponding parameters of the ASR-9, two other characteristics are markedly different. It utilizes a low peak power solid state transmitter that requires transmission of long, coded waveforms and a pulse compression receiver. Secondly, its pulse transmission sequence consists of short (five-pulse) bursts at both different pulse-repetition frequencies (PRF) and different RF frequencies. In this report, we assess the technical and operational issues associated with adding a WSP to the ASR-11. The existing WSP data processing and display technology are largely re-usable for the ASR-11 based WSP. Ground clutter filter coefficients and the length and number of coherent processing intervals would need to be changed to conform to the ASR-11 pulse transmission strategy, and straightforward adaptations to the equations used in the pulse-pair weather reflectivity and Doppler velocity estimation would be required. With these changes, the ASR-11 could host the WSP, subject to performance degradations for low reflectivity wind shear phenomena such as dry microbursts and gust fronts. A benefits assessment waas performed to evaluate the operational requirements for an ASR-11 based WSP. Given that the FAA has already committed to deploy improved Low Level Wind Shear Alert Systems (LLWAS) at most ASR-11 airports, the incremental safety benefits for the ASR-11 WSP appear to be less than the cost of the equipment. A case can be made for deployment based on "situational awareness" benefits that the WSP has been demonstrated to provide to air traffic controllers. We estimate that the value to the public and airline industry of reductions in aircraft delay, and avoidance of unnecessary diversions, would be in excess of eight million dollars per year tallied across 18 of the larger ASR-11 equipped airports.
Summary
The Federal Aviation Administration (FAA) is deploying a Weather Systems Processor (WSP) for the current-generation Airport Surveillance Radar - ASR-9. This modification exploits the coherency of the ASR-9 to perform Doppler wind measurement. Signature recognition algorithms then automatically detect low altitude wind shear phenomena, track thunderstorm motion and display appropriate...
READ MORE
The behavior of total lightning activity in severe Florida thunderstorms
July 1, 1999
Journal Article
Published in:
Atmos. Res., Vol. 51, Nos. 3-4, July 1999, pp. 245-265.
Summary
The development of a new observational system called LISDAD (Lightning Imaging Sensor Demonstration and Display) has enabled a study of severe weather in central Florida. The total flash rates for storms verified to be severe are found to exceed 60 fpm, with some values reaching 500 fpm. Similar to earlier results for thunderstorm microbursts, the peak flash rate preceeds the severe weather at the ground by 5-20 min. A distinguishing feature of severe storms is the presence of lightning 'jumps' -- abrupt increases in flash rate in advance of the maximum rate for the storm. The systematic total lightning precursor to severe weather of all kinds -- wind, hail, tornadoes -- is interpreted in terms of the updraft that sows the seeds aloft for severe weather at the surface and simultaneously stimulates the ice microphysics that drives the intracloud lightning activity.
Summary
The development of a new observational system called LISDAD (Lightning Imaging Sensor Demonstration and Display) has enabled a study of severe weather in central Florida. The total flash rates for storms verified to be severe are found to exceed 60 fpm, with some values reaching 500 fpm. Similar to earlier...
READ MORE
Understanding-based translingual information retrieval
June 17, 1999
Conference Paper
Published in:
4th Int. Conf. on Applications of Natural Language to Information Systems, 17-19 June 1999, pp. 187-195.
Topic:
R&D area:
Summary
This paper describes our preliminary research on an understanding-based translingual information retrieval system for which the input to the system is a query sentence in English, and the output of the system is a set of documents either in English or in Korean. The understanding module produces a meaning representation --- called semantic frame --- of the input sentence where the predicate-argument structure and the question-type of the input are identified, and each keyword is assigned its concept category. The translingual search module performs search on an English and Korean bilingual corpus tagged with concept categories. The results of our preliminary experiment, performed an a document set consisting of slides and notes from English and Korean briefings in a military domain, indicate that an understanding-based approach to information retrieval combined with concept-based search technique improves both precision and recall compared with a keyword match technique without understanding for both monolingual- and translingual retrieval. Current work is directed at further development of the system, and in preparation for tests on larger copora.
Summary
This paper describes our preliminary research on an understanding-based translingual information retrieval system for which the input to the system is a query sentence in English, and the output of the system is a set of documents either in English or in Korean. The understanding module produces a meaning representation...
READ MORE
Demonstration of a 630-GHz photomixer used as a local oscillator
June 13, 1999
Conference Paper
Published in:
1999 IEEE MTT-S Int. Microwave Symp. Digest, Vol. 1, 13-19 June 1999, pp. 127-130.
Topic:
Summary
We report the first successful demonstration of a photomixer local oscillator (LO) integrated with a superconducting heterodyne detector. The photomixer LO generated the difference frequency of two diode lasers by optical heterodyne conversion in low-temperature-grown GaAs. The measured receiver noise temperature, 331 K at 630 GHz, compares favorably with that achieved using a varactor-multiplier LO. We also report the design of a high-power distributed photomixer structure.
Summary
We report the first successful demonstration of a photomixer local oscillator (LO) integrated with a superconducting heterodyne detector. The photomixer LO generated the difference frequency of two diode lasers by optical heterodyne conversion in low-temperature-grown GaAs. The measured receiver noise temperature, 331 K at 630 GHz, compares favorably with that...
READ MORE