Query-by-example spoken term detection using bottleneck feature and Hidden Markov model

Query by example spoken term detection (QbE-STD) is an effective search mechanism to find spoken queries in spoken audio, especially for the source limited language. The dynamic time warping (DTW) algorithm is the state-of-art algorithm in this area. This paper presents some methods to improve the QbE-STD performance. First, the hidden Markov model (HMM) is adopted to model the keyword templates. Second, we apply the deep neural network (DNN) bottleneck feature to replace the traditional acoustic feature. The experimental results have demonstrated the effectiveness of the proposed methods. We are able to achieve an about 6% absolute improvement in F1 on a Tibetan corpus over the previous best baseline.