Mutitask Learning Based Muti-examples Keywords Spotting in Low Resource Condition

Keywords Spotting (KWS) is a critical task in speech recognition, aiming to spot the pre-selected keywords out of a continuous speech. In a typical low resource condition, we can only obtain dozens of examples of each keyword. How to make full use of multi-examples information to build an effective keyword spotting system is the present challenge since traditional keyword spotting technologies are not suitable. In this paper, we propose a multi-examples keywords spotting system, which gains a significant performance improvement by applying multitask learning technologies to extract feature and build model. A post-processing method is also used to reduce the false alarm rate.