The Study Of The Articulatory Mechanism Of Emotional Speech

The past research concerning emotional speech synthesis has been based on the analysis of the acoustic correlations of speech produced in different emotions. And the research has focused on English, Japanese, German and Swedish, less attention has been paid to Chinese emotional speech. Even though the way to express emotion is universal, the effect emotion imposes on intonational patterns is language-specific, especially in a language featuring lexical tones.The research on English shows that prosody can influence articulation. With the development of synthesis techniques, more and more attention has been paid to the effects of prosody. When speech is produced in different emotions, it is difficult to study its articulatory mechanism. Through EMA, this study recorded and analyzed the articulation of speech by a female speaker with neutral emotion. The speaker produced six monophthongs, including [a i u y e o], with four lexical tones under narrow focus. The effect of lexical tones on articulation was examined based on the measures of F0, F1, F2, F3.The first part of the thesis is the literature review. It shows only a few studies focus on Mandarin emotional speech.The second part of the thesis is the introduction of the way of recording with EMA and the way to preprocess the data, also introduce the vowel target algorithm.The third part is the analysis of the articulatory and acoustic data and the comparison between the two kinds of data.The fourth part is the analysis of the vowel space with the principal component analysis in Matlab, aiming at exploring the factors influencing the space. The results reveal that lexical tones have influence on articulation and for different vowels, the impact is different. For each vowel, when carrying different lexical tones, F1, F2, F3 are different. There is a kind of correlation between EMA data and acoustic data. Through the principal component analysis two factors are primary, one factor indicating the front-back dimension and the other indicating tongue height and the tongue shape.Finally, this part is the how to convert the EMA data to visualization model.