운율구 단위와 문장구조의 상관관계에 대한 실험음성학적 연구

Abstract

The current speech technology has been aiming to acquire much clearer and more natural synthetic speech sound. The naturalness of the two can be developed by an adequate phrasing of target sentence, of course, which seems to be strongly related to both syntactic and phonetic aspect simultaneously. The present study aims to describe, at one aspect, the relatedness between syntactic structure and prosodic phrasing through 2 different styles of speech data, viz. Read and Dialogue speech, respectively, and at the other, to establish a suitable phrasing pattern with respect to the purpose of acquiring more natural synthetic sound. The prosodic phrase, here, means a prosodic unit which can be clearly identified as having an evident break boundary at its final position in a sentence in the sense of both perceptual and acoustical viewpoint. The end of each prosodic phrase is, accordingly, marked as the point of major boundary in a sentence. In chapter 2, - I tried to classify each prosodic boundary as one of the 10 patterns which had been discriminated by the degree of durational lengthening or shortening and that of intonational rising or falling comparative to the reference mean duration which had been calculated from 156 read sentences. The Fo fluctuation was, however, calculated to a ratio comparative to the mean Fo value of each sentence.. This work has been carried out automatically by computation and the patterns were ordered from 0 to 9 according to its hierarchy. The 156 read and 100 dialogue speech sentences were, therefore, labelled automatically via such a tool. The text database of the former was tagged up to the level of grammatical relation, on the contrary, the latter was analyzed only to the naming of the grammatical morpheme information. There seems to be some difference between the two approaches at a glance but we can also find that two other strategies fulfill in common our working suggestion that we are now finding some correlation between grammatical information and prosodic phrasing. In chapter 3, 663 prosodic phrases in 156 read sentences were resulted from such a work. Among the 10 patterns, the highly ordered [0,1,2] labels were regarded as major prosodic boundary. I examined firstly the number of syllables within each prosodic phrase and secondary, the number of occurrence of the various grammatical relation pair when each of them posed at both sides of prosodic boundary with primary consideration, and also did the same work within the prosodic boundary with secondary importance. Lastly but not inferior to the above, the frequency of entire syntactic string within each prosodic phrase was ranked according to the hierarchical order. This made to know the fact that the modifiers including adnoun, adverb can be highly ranked as a prosodic unit which can solely function within a sentence. The pairs including subject*adnoun, predicate*subject, predicate*object, predicate*adverb, predicate*adnoun, locative*subject, locative*adnoun, adverb *adnoun were easily divided by the prosodic boundary, all of which can be referred from the tables 6-8. In chapter 4, the prosodic boundary information of travelling domain dialogue speech was labelled out with a principle that among the major boundaries selected by the above criterion, only those having silent interval should be regarded as the proper one. The statistic analysis was focused only to the grammatical informations paired each other at both sides of the major boundary. I presented all the cases of such pairs with the number of occurrence in the table 11. Having established 4 kinds of acoustic features, I arranged the duration and Fo ratio of penultimate, boundary, and post-boundary syllable along the X-Y two dimensional scale. Table 13 shows all of these. From the present discussion, most of the grammatical pairs seemed to have a characteristic that the intonation was tending to fluctuate from rising to falling and, simultaneously, the duration showed of a short-long-short or a long-short-short pattern in the same syllable string of penultimate-boundary-post_boundary. In order to make a work like this way fruitful, the level of natural language processing should be enhanced much more relative to current state. Only after the final result of NLP can assure of the grammatical correctness of anlayzed data better than what we imagine now, we can proceed more above.