Query By Example Spoken Term Detection Using Parallel Tokenizers
Spoken term detection refers to method of identifying the occurrences of words in an audio file. A technique
in which Query input is given in the form of human utterance is called Query By Example Spoken Term Detection. Spoken
term detection system employs Tokenizer for generating Posteriorgrams from audio speech. This paper presents an approach
where multiple Tokenizers are being used in parallel to enhance performance of STD system. Query and test utterances
were converted into query and test Posteriorgrams using theses Tokenizers and matched using dynamic time warping
method. Unsupervised Spoken term detection is considered under a scenario ,where zero prior knowledge or less knowledge
is available about the language.
Keywords— Query by Example; keyword spotting, Spoken Term Detection, Dynamic time Warping, Feature Extraction.