Chinese Prosodic Phrases Based On Text And Phonetic Features Boundary Prediction

At present,speech synthesis technology in People's Daily life and work can be seen everywhere,such as: public information consulting services and release,voice response of computer assistant,the mobile intelligent voice response service,television broadcast presided over the relevant text proofreading,daily communication support for the disabled,etc.With the popularity and application of speech synthesis technology people are increasingly concerned about the quality of speech synthesis,which affect the quality of speech synthesis of two important factors are the intelligibility and naturalness.At this stage,the intelligibility of speech synthesis has reached the ideal level,but the naturalness also restricts the quality of speech synthesis.The main problem with the existence of speech synthesis of the expression of the lack of feelings like the same people,the rhythm of the sentence is poor,the machine is too heavy.Therefore,the main work of this paper is to study the prosody structure of Chinese,from the perspective of prosodic phrase boundary to improve the naturalness of speech synthesis.Therefore,the realization of the right rhythm segmentation,correctly grasp the rhythm of discourse structure,is to strengthen the synthetic speech rhythm,the key to improve the naturalness.The prosodic boundary prediction problem plays an important role in improving the naturalness degree of speech synthesis.In order to improve the natural degree of synthetic speech,the main work of this paper is to study the prosodic structure of Chinese to realize the automatic prediction of prosodic phrase boundary.The main contents of this paper:(1)Analysis and Extraction of Text and Phonetic characteristics at the Boundary of Prosodic Phrases.At present,for most researchers,most of the words part of speech,word length and other representative text features are used on the rhythm phrase boundary prediction,However,it is not ideal to only use the textfeature for prosodic phrase boundary prediction.Therefore,this paper proposes a Chinese prosodic phrase boundary prediction method based on the combination of text and speech,By using the feature selection algorithm of different models and analyzing the text and speech features at the boundary of prosodic phrases,the optimal feature set of the model is selected to train and forecast the model.(2)Rhythm phrase boundary prediction based on text featureFrom the manual annotation corpus to extract text characteristics,The text characteristic mainly contains:Word,part of speech,word length and other common text features.By analyzing the atomic characteristics,similar compound features and different kinds of complex features in the conditional random field model,so as to determine the optimal feature template of the model,and construct the rhythm phrase boundary prediction model.At the same time,the prosodic phrase prediction model is established by using the maximum entropy method.By comparing the two results,the automatic prediction of the rhythm phrases of Chinese is realized.(3)Rhythm Boundary Prediction Based on Text and Speech CombinationOn the basis of the text feature,the phonetic features are added,including the syllable extension ratio,the final length extension ratio,the syllable type,the relative length of the silent segment,the type of tone and so on.Through the analysis of the extraction feature,the optimal feature selection algorithm is selected,and then the different entropy phrase boundary prediction system is established by using the maximum entropy and the conditional random field method,and the experimental results based on the text feature are compared and analyzed.