In the evaluation of NIST Open Keyword Search 2014 (OpenKWS14), the system achieved a good performance, researched and developed by the
MSIIP members Zhipeng Chen, Teng Zhang and Tianying Chen.
OpenKWS is an international authority evaluation in the field of key word spotting (KWS) organized by National Institute of Standards and
Technology (NIST), requiring high-performance speech key word systems constructed in limited time with limited data resource for a
unknown language beforehand. This year, paticipants included 4 team supported by the IARPA Babel project (leaded respectively by
IBM, BBN, CMU and ICSI) and other voluntary research institutes.
The IARPA Babel project started in 2011. There has been an OpenKWS evaluation organized by NIST since 2013 and it was the second term this
year. MSIIP was the first time to participate, and achieved more than 0.5 (one of 4 teams) in the ATWV index for the FullLP task (60-hour training data
with labelling), beyond 0.3 (one of 5 teams) in the ATWV index for the LimitedLP task (10-hour training data with labelling).
OpenKWS14 Workshop was held in Washington DC on July 8th, 2014. Associate prof. Ji Wu, the director of MSIIP, attended this conference and
made a report named MSIIP OpenKWS Systems and Analysis as team summary in PI meeting, on behalf of MSIIP.