Research On The Pronunciation Method In Both Chinese And English Based On DIVA Model

DIVA(Directions Into Velocities of Articulators) model is Neural computing model, with speech production and acquisition capability, takes 29 basic phonemes in English as research background. However, it makes the model not adapt to other languages and then brings limitation. This paper aims to expand the model theory to other languages, such as Chinese pronunciation that is more complicated than English.Due to the big difference of the number of phonemes and pronunciation brain mechanism between Chinese and English, the original model only applies to the brain mechanisms of English speakers and produces English vowels. And the original training and learning algorithm of the model doesn’t recognize Chinese pronunciation well. Therefore, this paper improves the original and makes it suitable for Chinese pronunciation.At first, this paper introduces the basic principle and two learning stages of DIVA model. And then it explores the acoustics characteristic about Chinese compound vowels and shows the trend chart of their formant frequency. Based on the known identification of English consonant /r/ and /l/ and English vowels auditory perception space theory, this paper studies the pronunciation and perception space about Chinese pronunciation in depth. The learning algorithm of DIVA model is improved and the training parameters are adjusted and hybrid model and the formant extraction method(Linear Prediction and Mel Frequency Cepstrum Coefficient, LPMCC) are used. It proves that modified model can adapt to the pronunciation of Chinese signal vowels and compound vowels and pronunciation effect is excellent.This paper lays theoretical and practical foundation for applying DIVA model to speech production and acquisition about Chinese and expanding into other Chinese consonant sounds such as /zh/,/ch/,/sh/.