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

A phoneme recognition expert system combining spectrogram reading knowledge and neural networks is described. Consonant recognition is performed in three stages: (1) consonant segmentation based on spectrogram reading knowledge, (2) consonant identification based on neural networks, and (3) final consonant determination combining the results of the consonant segmentation and identification stages. The mechanisms for combining consonant segmentation and identification are studied to enhance their respective advantages. Consonant recognition experiments show that the organic combination of segmentation and TDNN improves not only phoneme identification performance but also segmentation accuracy. Furthermore, it effectively reduces insertion errors. Vowel recognition uses phoneme-spotting neural networks for vowel detection and spectrogram reading knowledge for verifying its identity and the boundaries. The effectiveness of this approach is shown through a vowel detection experiment