Researcher, Natural Language Understanding
Nuance Communications R&D,
1 Wayside Road, Burlington, MA, 01803
Tel: (M) 781-475-9546
Email: firstname.lastname@example.org email@example.com
Web page: http://home.comcast.net/~dutanico/
Ph. D. Computer Science Michigan State E. Lansing, MI 2000
M.S. Computer Science The University of Iowa Iowa City, IA 1996
M.S. Statistics University of Paris XI Paris, France 1992
B.S. Applied Mathematics University of Romania 1991
Researcher Nuance Communications 2005 - present
Scientist BBN Technologies 2000 - 2005
Research Assistant Michigan State University 1997 - 2000
Research Assistant Siemens Corporate Research 1997 - 1999
Teaching Assistant Michigan State University 1996 - 1997
Instructor The University of Iowa 1996
Teaching Assistant The University of Iowa 1994 - 1996
Research Assistant INRIA - Rocquencourt, France 1993
Research interests: pattern recognition and machine learning
techniques applied to vision, speech, language, biometrics, robotics and
Professional Responsibilities and Projects
At Nuance Communications Nicolae Duta has been a member of the Natural
Language Understanding R&D group. He has been responsible for the research,
design and development of the language modeling and understanding modules in
the Nuance Recognizer V9 and integrating the corresponding modules in the
OpenSpeech® Recognizer with those in the Nuance 8.5 Voice Platform. His
research of data sharing over multiple applications and domains contributed to
creating best training practices for the Nuance Directory Assistance 6.0 and
Nuance Call Steering products and is currently being patented [P2]. In the same
time, he continued to have an active role in the computer vision and biometrics
research communities by publishing technology survey papers , delivering
invited talks [T1] and serving on technical committee boards [B].
Between 2000 and 2005 he was a scientist at BBN Technologies where he has
done research and development in natural language modeling and speech
recognition. He developed language models for automatic speech to text
translation in both broadcast news and conversational telephony domains and three
languages (English, Mandarin and Arabic)[6-7]. He designed a full language
modeling package which can handle billion-words corpora [T2] which helped the
BBN team win the Best Speech Recognition System award in the April 2003
DARPA evaluation. He also qualitatively analyzed the errors made by current
state-of-the art large vocabulary continuous speech recognition systems as well as
the errors made by the human annotator(s) who are used to establish the ground-
truth against which speech recognition systems are evaluated . The error
analysis produced a lower bound on the word error rate that can be achieved in
practice as well as suggested solutions for building an online real-time LVCSR
Between 1997 and 2000 he was a research assistant in the department of Computer
Science and Engineering at Michigan State University. He developed trainable
algorithms and implemented systems for object (medical structures [3,6,7], human
faces , human hand shapes and palm-prints [1,4], vehicles ) learning and
retrieval in digital images.
In 2000 he developed a very fast approximate string matching/aligning algorithm
that can efficiently handle word mutations (different word spelling due to letter
substitution, deletion or insertion).
During the summer months of 1997-1999 he was a research assistant in the
Imaging and Visualization Laboratory, at Siemens Corporate Research, Princeton,
NJ where he worked on left ventricle detection and segmentation in MR cardiac
images [4,9]. That work is patented and is part of Siemens’ commercial image
analysis package called Argus.
Between 1994 and 1996 he conducted research on MR brain image segmentation
[7,14] in the College of Engineering Imaging Group at the University of Iowa.
In 1993 he was part of the research team in the Syntim Laboratory at the Institut
National de Recherche en Informatique et en Automatique (INRIA),
Rocquencourt, France where he worked on road detection in SPOT satellite
Products and systems developed (demos available on request,
see my web-page for more details)
2005 - 2009 Nuance Communications: Language modeling and
understanding modules in Nuance Recognizer V9
• Used in real-time commercial speech recognition and call routing.
Written in C/C++.
2002 - 2004 BBN Technologies: Language modeling toolkit
• Handles billion-word data, multiple discounting models, on-the-fly word
compounding and computes the word transition probabilities in almost
real time on a Linux PC platform. It is written in C and can be compiled
on most platforms, the 64-bit Intel version is most efficient.
2000 - 2002 BBN Technologies: Rejection sub-system of the BBN
automated directory assistance system
• Rejects to a human operator the directory assistance calls which cannot
be automatically processed. Written in C and Perl.
1998-2000 Siemens Corporate Research: Left ventricle detection in MR
images (part of Simens ’ Argus)
• Real time ventricle detection and display, handles multiple ventricle
slices and imaging parameters. Written in C with a Matlab GUI.
Commercialized as part of Siemens’ MR software.
1998 - 2000 Michigan State University: Automatic shape learning system
• Clusters a set of given object shapes, discards outlier shapes and
computes the average shape (model) and modes of variation for each
shape cluster. Used for shape analysis and retrieval. Written in C with
Matlab GUI and demo, has been used for research at several universities.
1997 - 2000 Michigan State University: Human face detection system
• Face detection in black and white pictures, 85% correct detection rate,
10^(-5) false accept rate. Written in C.
2000 Michigan State University: Palmprint-based ID verification system
• Written in C and Matlab.
1999 Michigan State University: Hand shape-based ID verification system
• Written in C and Matlab, tested on 50 subjects, 2% ID verification error.
1998 Michigan State University: Corpus Callosum segmentation system
• Segments in almost real time the corpus callosum in MR images. Written
in C and Matlab.
1995 - 1997 The University of Iowa: Neuro-anatomical brain structure
• Segments 10 structures (shown in the brain figure above), able to
automatically detect and repair outlier boundaries. Written in C with a
1993 INRIA - Rocquencourt, France: System for road detection in
• Completely automatic highway detection/tracking in huge SPOT images.
Written in C with a Motif GUI.
Programming experience, languages and platforms
Developed and implemented advanced vision, speech, language and other
pattern recognition and AI algorithms since 1990.
• Pascal and C since 1988, taught C programming at the University of
• Prolog since 1989, used it for solving AI problems
• Motif and OpenLook since 1991, used them for some system GUIs
• Matlab since 1995, used it for most system GUIs and numerical analysis
• C++ since 1996, conducted laboratory sessions at Michigan State
• C-shell since 1996, conducted laboratory sessions and automated project
grading at Michigan State University (1997)
• Perl since 2000, used it for almost all string processing needed in speech
• Small programs in Java and HTML
• SunOS/Solaris since 1991
• IBM PC/Windows98,NT,2000 since 1996
• IBM PC/Linux since 1999
Technical Program Committee Member
• International Conference on Pattern Recognition, Cambridge, UK, 2004
• International Conference on Pattern Recognition, Hong Kong, 2006
• Biometrics Symposium at The Biometric Consortium Conference (BCC)
Member, IEEE and IEEE Computer Society
Reviewer: IEEE Transactions on Pattern Analysis and Machine Intelligence,
IEEE Transactions on Multimedia, Pattern Recognition, Pattern Recognition
Letters, Computer Vision and Image Understanding, Pattern Analysis and
P1. System and method for segmenting the left ventricle in a cardiac image,
United States Patent 7,400,757, 2008.
P2. Speech Recognition Semantic Classification Training, United States Patent
Application 2639B67, 2008
P3. Method for learning-based object detection in cardiac magnetic resonance
images, United States Patent Application 20030035573, 2003.
P4. Deformable Matching of hand shapes for verification, United States
provisional Patent, 1999.
Cambridge, MA, 02138 East Lansing, MI, 48824
Phone: 617-873-3332 Phone: 517-355-9282
Fax: 617-873-2473 Fax: 517-432-1061
E-Mail: firstname.lastname@example.org E-Mail: email@example.com
Professor Milan Sonka
Dept. of Electrical & Computer Professor Sridhar Mahadevan
Engineering Dept. of Computer Science
The University of Iowa University of Massachusetts
Iowa City, IA, 52242 Amherst, MA 01003
Phone: 319-335-6052 Phone: 413-545-3140
Fax: 319-335-6028 Fax: 413-545-1249
E-Mail: firstname.lastname@example.org E-Mail: email@example.com