Research On The Pronunciation Mechanism Of Chinese Vowel Tone And English Stress Based On DIVA Model

As the speech technology continues to develop, there are lots of speech production model ofhigh quality, but they don’t involve the principles and applications of neurophysiology. In recentyears, the main ideas of the artificial speech synthesis system to pursuit is that description andsimulation of the human brain function involved in speech understanding and generation region inneuropsychology and neuroanatomy level. The DIVA (Directions Into Velocities of Articulators)model reported by professor Guenther (Frank H. Guenther) of speech lab at Boston Universityreflects the correlations between neuroanatomy data and brain area. It’s goal is to generate words,syllables or phonemes by controlling an analog channel, besides it is the most thorough of speechacquisition and generation model which have biological significance. However, DIVA model takesthe English speech acquisition and generation as research background, and does not cover thediscussion of speech prosodic features. As we know, prosodic features include tone, intonation andstress both in Chinese and English. On linguistics, Chinese tone has been conclusively linked to theidentification of words and sentences. English stress could differentiate the meaning of a word andthe part of speech. This paper studies the means, method and process of Chinese vowel tone andEnglish stress pronunciation based on DIVA modelFirst, this paper introduces the basic concept, structure and present situation of the DIVA model,the rhythm of Chinese tone and tone-related acoustic parameter. Then it uses the pitch scalingfunction to work out the tone-related track of pitch, and modifies the movement command of DIVAmodel through the TD-PSOLA to achieve the goal of Chinese vowel tone pronunciation. The newDIVA model could simulate the neural correlates of Chinese vowel tone. Compared with the clinicaloutcome, it proves the correctness and applicability of the improved model.Second, the rule of English stress and the stress-related acoustic parameters are studied. Also thepreprocessing of speech signal and the extraction protocols of pitch and duration are discussed.Then make use of the TD-PSOLA to modify the movement command of the DIVA model andachieve the goal of English stress pronunciation.Finally, this paper uses the test of pitch and duration compensations to validate the correctnessof the improved model through pitch perturbation experiments. This paper lays theoretical and practical foundation for building the “Neuralynx System” withthe thinking characters of Chinese, and provides relevant basis of the research on neural correlatesof speech prosody.