SUNGBOK LEE
Tel: 213-821-2721,
FAX: 213-740-9306
Email: sungbokl@usc.edu
RESEARCH INTERESTS
• Speech Production Modeling
• Automatic Speech Recognition
• Speech and Language Processing for Human-Machine Interactions
Research Assistant Professor,
·
Conducting
various researches on speech production and speech production modeling using
fast magnetic resonance imaging (MRI) and electromagnetic articulograph (EMA).
Developing efficient and robust speech processing technologies for automatic
pronunciation evaluation and emotional speech recognition/synthesis.
·
Guiding two
PhD-level research groups at the Speech Analysis and Interpretation Lab (SAIL)
for automatic pronunciation evaluation and emotional speech
recognition/synthesis for the use in spoken dialogue management systems.
Consultant, Lucent Bell-Labs,
·
Responsibility
included the acoustic modeling and dialogue management system development for
the Lucent Bell Lab’s DARPA Communicator project.
Consultant, AT&T Labs - Research,
·
Responsibility
included acoustic and language model training and testing, performance
optimization of the AT&T Watson recognition engine for the AT&T DARPA
Communicator project. Learned a script language
for mixed-initiative dialogue system design and enhanced the functionality of
the Communicator dialogue system system.
·
Contributed to
achieve an outstanding performance in the June-2000 DRAPA evaluation period.
·
Contribute to the
analysis of DARPA Communicator evaluation dialogue corpus for prediction of
problematic dialogues and for development of objective performance measure of
dialogue management system.
Consultant, Lucent Speech Solutions,
·
Acquired working
knowledge and experience in most aspects of large vocabulary, speaker
independent automatic speech recognition system including front-end acoustic
processing, acoustic model training and testing including garbage model,
performance optimization, and grammar development.
·
Trained and
tested acoustic models and context-free grammars for the Lucent ASR engine.
·
Managed the
integration process of Lucent speech recognition
engine with UniSys' “Natural Language Speech
Assistant” speech application development tool.
·
Tested and
analyzed the performance of the Lucent ASR engine for accuracy, memory usage,
and response-time with various grammars.
Visiting Scientist,
·
Planned and conducted a research on the
developmental changes of temporal and spectral parameters of speech in
conjunction with the improvement of children's speech recognition. The study was published in the Journal of
Acoustical Society of America as a selected research article.
Research Engineer, Central Institute for the Deaf,
·
Successfully
designed and implemented data collection programs for creation of children's
speech databases from 365 children using microphone and telephone.
·
Investigated
performance and difference in vowel classifications by human listeners and
objective vowel classifiers including the neural network.
·
Developed a
time-domain pitch estimation method based on the LPC residual error and demonstrated
its effectiveness on children’s voice and diplophonic
voices. The method was also successfully
applied for the study of characteristics of voicing between deaf and normal
children.
·
Conducted a study
on formant frequency variability in the vocal tract as a function of the degree
and location of the tongue constriction for the examination of quantal theory
of speech production
·
Developed an
arbitrary spectral envelop synthesizer of vowel-like sounds for speech
perception research.
COMPUTER SKILLS:
Language: C/C++, Perl
Operating System: Windows family, Unix,
Linux
Scientific Software: Matlab,
SPSS, Systat
Ph.D., Biomedical Engineering,
M.S., Physics,
B.S., Chemistry,