Combining phoneme identification neural networks into an expert system using spectrogram reading knowledge

Abstract

This p a p e r d e s c r i b e s a p h o n e m e recognition e x p e r t system combining spectrogram r e a d i n g k n o w l e d g e a n d neural networks. Consonant recognition is performed i n t h r e e stages: 1) C o n s o n a n t s e g m e n t a t i o n b a s e d on spectrogram reading knowledge. 2) Consonant identification based on neural networks. 3) Final consonant determination combining the results of the consonant segmentation a n d identification stages. T h e m e c h a n i s m s f o r c o m b i n i n g consonant segmentation a n d identification were studied to en hence their respective advantages. Consonant recognition experiments a r e carried out, a n d show t h a t t h e organic combination of segmentation a n d TDNN improves not only phoneme identification performance b u t also segmentation accuracy. Furthermore, it effectively r e d u c e s i n s e r t i o n errors. Vowel recognition uses phoneme-spotting n e u r a l networks for vowel detection a n d s p e c t r o g r a m r e a d i n g knowledge for verifying its identity a n d t h e boundaries. The effectiveness of this a p p r o a c h is shown through a vowel detection experiment.

On the use of neural networks for automatic vowel recognition

“…In recent years interest in ANN for processing speech patterns and obtaining ASR has been generated widespread research activity [S] [9] [ IO][ 171.…”