The objective of the Open Keyword Search Evaluation (OpenKWS) evaluation series is to support research in, and help advance the state of the art of, keyword search (KWS) technologies - technologies that locate a specified, potentially multi-word keyword in the specified language. The goal of the program is to reduce the difficulty of building high-performing KWS systems on a new language quickly with diminishing data resources and time.
The OpenKWS project is an extension of the 2006 Spoken Term Detection evaluation, which tested KWS systems on English, Mandarin, and Arabic recordings from Broadcast News, Conversational Telephone Speech (CTS), and conference meeting data. OpenKWS is a multi-year effort to test KWS systems on a new "Surprise" language each year. Participants will receive conversational telephone speech training, development, and testing resources each year for the language, but have a limited time to build their systems. At the conclusion of the evaluation, participants will be invited to an evaluation workshop held in the Washington DC Metro Area to discuss the results of the evaluation.
Past Surprise languages were Vietnamese, Tamil, and Swahili for 2013, 2014, and 2015 respectively. The most recent OpenKWS evaluation, OpenKWS16, took place in the summer of 2016.
openkws-poc [at] nist.gov