DocumentCode :
2747755
Title :
Neural network models for spotting stop consonant-vowel (SCV) segments in continuous speech
Author :
Sekhar, C. Chandra ; Yegnanarayana, B.
Author_Institution :
Dept. of Comput. Sci. & Eng., Indian Inst. of Technol., Madras, India
Volume :
4
fYear :
1996
fDate :
3-6 Jun 1996
Firstpage :
2003
Abstract :
Spotting subword units in continuous speech is important for realizing a task independent and vocabulary independent continuous speech recognition system. In this paper we consider different neural network models and architectures for spotting subword units belonging to the confusable set of stop consonant-vowel (SCV) classes in Indian languages. In the proposed approach for spotting SCV segments, the vowel onset points (VOPs) in continuous speech are located using a neural network model. Neural network classifiers trained with the SCV data excised from continuous speech are then used to scan the speech segments around VOPs for spotting SCVs. In our studies we consider the one-class-one-network (OCON) and all-class-one-network (ACON) architectures using multilayer perceptron and time-delay neural network models for classification of SCVs. Spotting performance of these models and architectures is illustrated for frequently occurring ten SCV classes
Keywords :
multilayer perceptrons; pattern classification; speech recognition; Indian languages; all-class-one-network; continuous speech; multilayer perceptron; neural network classifiers; neural network models; one-class-one-network; stop consonant-vowel segments; subword units; time-delay neural network; vowel onset points; Artificial neural networks; Computer architecture; Computer science; Delay effects; Feedforward systems; Hidden Markov models; Intelligent networks; Neural networks; Speech recognition; Vocabulary;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Neural Networks, 1996., IEEE International Conference on
Conference_Location :
Washington, DC
Print_ISBN :
0-7803-3210-5
Type :
conf
DOI :
10.1109/ICNN.1996.549209
Filename :
549209
Link To Document :
بازگشت