The Signal, Speech and Language Interpretation Laboratory
The University of
Washington, Seattle is home to a broad community of speech and
language research, within which the SSLI Lab is a cornerstone of the
technology efforts. The lab was founded in 1999 and currently supports
a group of 4 senior researchers and about 25 graduate students engaged in
solving the challenging problems in speech and language technology and
general signal interpretation.
Principal Researchers
Collaborators
Due to the interdisciplinary nature of speech and language processing, SSLI researchers collaborate with colleagues both within and outside our department, notably:
In addition, we have collaborations with researchers at several other institutions, including: ICSI, SRI, ISI, Columbia, NYU, UMd/Purdue, Stanford, USC, and the University of Karlsruhe.
Research Areas
- dynamic acoustic and language models for large vocabulary speech recognition
- modeling uncertainty and leveraging structure in speech (e.g. segmentation, prominence) in spoken language processing
- multilingual speech processing
- natural language processing and machine translation
- language technology for education applications
- speech-based human-computer interfaces
-
low-power and efficient software algorithms - optimize at the software
level to reduce power consumption for speech systems running on portable
devices
-
incorporating ASR confidence in information extraction from speech
-
prosody and dialog tracking - integrating prosodic and language cues for
recognizing dialog acts and topic structure.
-
graphical models for speech and language technology - adjust the models automatically to better match
the properties found in very large databases.
Programmatic Goals
Our overall goal is to maintain a stimulating research and teaching environment
where participating members have expertise in a broad range of technical
areas. This will facilitate the multidisciplinary research necessary to
successfully attack the most challenging problems in speech processing.
In particular, we intend:
-
to explore fundamentally new methods in speech and language technology,
and
-
to produce an educational environment where students quickly become proficient
in a wide range of speech technologies.
Students
We seek students with backgrounds in electrical engineering, computer science or linguistics, with an interest to learn about the other disciplines. Students can expect to develop expertise in a wide variety of statistical modeling techniques as well as software and systems engineering skills. They also have access to courses and potential collaborations with faculty in related departments on campus, including EE, CSE, Statistics, Linguistics, Speech and Hearing Sciences, and Psychology. We place a high value on oral and written communication skills, and students have the opportunity to give presentations in lab seminars as well as at international conferences. Many students take a summer or a term in the academic year to do internships at leading research labs. SSLI graduates have taken positions in industry as well as academia, depending on their career goals and interests.
Facilities
The SSLI is supported by a large computer cluster with 7 file servers and more than 120 compute servers, mostly linux-based. Efficient job scheduling is based on condor. We provide standard speech and signal processing tools, as well as develop our own, including GMTK.