paulb@ttidca.TTI.COM (Paul Blumstein) (12/02/89)
I have heard rumors that somewhere in Great Britain there has been a major
breakthrough in the field of Speech Recognition. However, I have not
heard anything except for vague references.
Is there someone on usenet that knows anything about this or can point me
to a reliable source of information? If you feel that your answer is of
general interest, you may post it here. If not (or you are too shy to
post), please send me e-mail. If I get responses that I feel may interest
the community, I will post a summary.
Thank you, in advance.
=============================================================================
Paul Blumstein | TB, or not TB, that is the congestion. Consumption be
Citicorp/TTI | done about it? Of cough, of cough. -- Woody Allen
Santa Monica, CA +-------------------------------------------------------
{philabs,csun,psivax}!ttidca!paulb or paulb@ttidca.TTI.COM
DISCLAIMER: Everything & everyone is hereby disclaimed!danforth@riacs.edu (Douglas G. Danforth) (12/07/89)
]
]I have heard rumors that somewhere in Great Britain there has been a major
]breakthrough in the field of Speech Recognition. However, I have not
]heard anything except for vague references.
]
]Is there someone on usenet that knows anything about this or can point me
]to a reliable source of information? If you feel that your answer is of
]general interest, you may post it here. If not (or you are too shy to
]post), please send me e-mail. If I get responses that I feel may interest
]the community, I will post a summary.
]
]Thank you, in advance.
]=============================================================================
]Paul Blumstein | TB, or not TB, that is the congestion. Consumption be
]Citicorp/TTI | done about it? Of cough, of cough. -- Woody Allen
]Santa Monica, CA +-------------------------------------------------------
]{philabs,csun,psivax}!ttidca!paulb or paulb@ttidca.TTI.COM
]DISCLAIMER: Everything & everyone is hereby disclaimed!
---------------------------------------------------------------------------------
While attending the recent NIPS (Neural Information Processing Society)
conference in Denver, Colorado, U.S.A, I had the good fortune to be able to speak
to John Bridle of Royal Signals and Radar Establishment in Malvern, England
who works with Roger Moore who was mentioned on the CNN television boadcast.
Dr. Bridle's response was simply that their system "ARMADA" was really
very similar to the SPHINX system created by Kai-Fu Lee at Carnegie-Mellon.
I believe the press picked up on the words "first of its kind" in the press
release and missed the phrase "to be developed in the UK" which lead to the
use of "breakthrough" in the CNN report.
John was kind enough to furnish me with a copy of the press release which
I reproduce below. For those not familiar with SPHINX I suggest you take a
look at the Carnegy-Mellon technical report: "Large-Vocabulary Speaker-Independent
Continuous Speech Recognition: The SPHINX System" by Kai-Fu Lee, April 18, 1988
CMU-CS-88-148.
Douglas Danforth
danforth@riacs.edu
-------------------------------------------------------------------------------
SPEECH RESEARCH UNIT
Royal Signals and Radar Establishment
St. Andrews Road
Malvern, Worcs.
WR143PS
0684-894074
PRESS RELEASE
18 September, 1989
RSRE Speech Research Unit Demonstrates Large-Vocabulary
Continuous Speech Recognition
The Speech Research Unit at RSRE Malvern has announced the 'ARMADA' system
which demonstrates large-vocabulary continuous speech recognition. The
system, which currently runs in near real-time on an array of 6
transputers, is the first of its kind to be developed in the UK and is
technically equivalent to advanced US systems developed under the Spoken
Language Systems component of the DARPA Strategic Computing Initiative.
'ARMADA' is the culmination of an intensive 9-month research project at
RSRE and is the latest demonstration of advanced speech technology at the
Speech Research Unit. The SRU carries out a programme of generic research
into speech recognition and synthesis and has contributed directly to
earlier commercial speech recognition systems such as LOGOS from Logica,
ASR1000 from Marconi and MICROTALK from the British Technology Group.
The 'ARMADA' system demonstrates high-accuracy recognition of continuously
spoken airborne reconnaissance reports. This task involves well defined
voice procedures which are typically of many potential civil and military
applications. In speaker dependent evaluations on a 500 word vocabulary
and in the absence of any grammatical constraints the recogniser scores an
impressive 82% average word accuracy. This figure rises to 99% when a
grammar is used to aid the recognition process.
The advantages of the 'ARMADA' system stem from the use of advanced
statistical techniques (based on hidden Markov modelling) to develop
accurate context-sensitive models of the acoustic realisations of phonemes
which are the building blocks of speech. This means that, unlike previous
systems, 'ARMADA' can recognise words which do not occur in the training
data, needs minimal speaker enrolment and can easily be reconfigured for
new applications. These are all key requirements for practical
large-vocabulary speech recognition.
'ARMADA' presents a unique UK opportunity for assessing the potential of
state-of-the-art speech recognition. In addition, 'ARMADA' is a valuable
tool for speech technology research, providing a testbed for novel
algorithms and a vehicle for the integration of speech and natural language
processing.