Loading...

Discriminative Articulatory Models for Spoken Term Detection in Low-Resource Conditions

Gomar, Zahra | 2015

540 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 48017 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Sameti, Hossein
  7. Abstract:
  8. This thesis is focused on the spoken term detection system based on speech recognition in low resources conditions. A spoken term detection system is composed of two parts: speech recognition and search. In search of words, the method of proxy words is used as a basic approache to overcome the problem of OOV words. The main challenge in this thesis in the context of low resources, is poor training acoustic and language models and the small lexicon in speech recognition. Small lexicon increases the number of OOV words. In this thesis, two innovation has been proposed to improve the basic system. The first is training a bottleneck neural network for extraction the articulatory features of phonemes and using them in speech recognition to improve the acoustic model. This improves the performance of spoken terms detection systems in terms of low resources conditions. In this way, 37.7% relative improvement in MSE of articulatory network and from 0.2 to 5 percent absolute improvement in PER of phoneme speech recognition with different acoustic models and from 16 to 470 percent relative im-provement in ATWV of search in the phoneme lattice were obtained, but this method did not result in im¬provemen¬t in word speech recognition. The second was using an expansion lexicon method to improve detection of OOV words. In this way, from 1.78 to 2.9 percent absolute improvement in WER of word recognition and 49/8 percent relative improvement in ATWV of the search section were resulted
  9. Keywords:
  10. Speech Recognition ; Spoken Term Detection ; Out of Vocabulary Words ; Discriminative Model ; Low Resource Condition ; Articulatory Features

 Digital Object List

 Bookmark

No TOC