Multilingual Bottleneck Features for Query by Example Spoken Term Detection

State of the art solutions to query by example spoken term detection\n(QbE-STD) usually rely on bottleneck feature representation of the query and\naudio document to perform dynamic time warping (DTW) based template matching.\nHere, we present a study on QbE-STD performance using several monolingual as\nwell as multilingual bottleneck features extracted from feed forward networks.\nThen, we propose to employ residual networks (ResNet) to estimate the\nbottleneck features and show significant improvements over the corresponding\nfeed forward network based features. The neural networks are trained on\nGlobalPhone corpus and QbE-STD experiments are performed on a very challenging\nQUESST 2014 database.\n