Abstract
This software is used for multi-lingual (Czech and English) search for relevant words or short phrases in the archive of Holocaust survivors, managed by USC (University of Southern California) Shoah Foundation Institute (http://dornsife.usc.edu/vhi/), which contains more than 110,000 hours of records in 32 languages, with approximately half of these interviews is conducted in English. Czech part of the archive accounts for approximately one thousand hours.
For the purposes of searching in the system MCLAAS are Czech and English speech data first processed with the appropriate speech recognition module (SEASR-CZE - see http://www.kky.zcu.cz/en/sw/SEASR-CZE or SEASR-ENG - see http://www.kky.zcu.cz/en/sw/SEASR-ENG) and then a so-called index is created, which is a machine representation of recognized utterances, which speeds up the search for a desired word or phrase. Both retrieval systems currently operate only with an index based on word representation - phonetic search will be implemented later.
Each system is currently looking for occurrences of words or phrases in about 1000 hours of video. Those are all data available in the case of Czech; in English there are more than 50,000 hours, but the recognition and indexing of all this volume will require parallelization of individual processes.
Cross-searching (query in English, data / interviews in English and Czech) in the system is facilitated by automatic query translation. Implementation was carried out as a special version of MTMonkey (http://ufal.mff.cuni.cz/mtmonkey).