Template-based Keyword Search with pseudo posteriorgrams
Batuhan Gündoğdu, Leda Sarı, Gözde Çetinkaya, Murat Saraçlar · 2016
In this work, a template-based search approach is adopted for the Keyword Search (KWS) problem on two of the low-resource languages (Turkish and Swahili). In low-resource languages, the use of Large Vocabulary Continuous Speech Recognition (LVCSR) systems in KWS tasks may perform poorly especially on out-of-vocabulary words. In the proposed method, the keywords are modeled to be in the same form of the audio search document in an artificial manner and utilizing template-based search methods, the performance of the baseline system is improved.