| With the rapid development of speech technologies, speech engineers are requiring more thorough prosodic details to build a better prosodic model and to improve the naturalness of synthesis speech.With this aim, a series of experiments are carried out to investigate how pitch accent and prosodic boundary are interacted in natural speech. We shall first examine how the pitch accent words distribute at prosodic phrases; then we will be concerned with the way in which the pitch of preboundary words under a variety of structures is affected by their associated rising boundary tone; and at last we will check how high pitch accent words are aligned with their related syllables in the context of falling tones.A large-scale American English corpus is used in the research. In the paper, we present a method to label this natural speech corpus from a perspective comparable to Chinese prosody labeling.We report the differences among the three-tier breaks: Break 4, 3, 2, and their relationship with the boundary tones and the pitch accent word immediately ahead of the breaks. The tones on Break 4 and Break 3 behave similarly, while the length of silence period after the two categories of breaks are significantly different. Both the speaker's upward pitch resetting and the ratio of the pitch accent word immediately ahead of the breaks are increasing with the level of breaks going down. At the end of Chapter 3, we discuss the differences between our labeling system and ToBI, and also the prosodic differences between Chinese and English.To investigate the interaction of the pitch of preboundary words and the rising boundary tone, we check words under a variety of structures. The materials are chosen from monosyllable, bi-syllable and tri-syllable words that are immediately ahead of Break 3 and Break 4 with rising boundary tone. These words are classified according to the place of lexical stress of the word, and are further separated into nuclear and non-nuclear categories. With or without silence period at the break is also one of the dependent factors.We notice under this environment, nuclear words usually have the rise-fall-rise pitch movement. Three points are studied, the two turning points of the rise-fall-rise pitch contour and the last point. We name the three points "peak", "valley" and "end". The information of lexical stress is very clear when the word is a pitch accent. They have salient peaks at the lexical stress of the word. This peak is even higher than the final point of the rising pitch. However, preboundary non-nuclear words generally have a simple rising pitch contour. Because the lexical stress is not the sentence stress, the rising boundary tone takes control at the beginning of the word. This happens to all branches with some of monosyllable pitch accent words as exceptions. The special pitch representation in monosyllable words might be because the single syllable is very apt to previous prosodic environment.Nuclear words before Break 3 have higher lexical pitch peak than that of Break 4. This might be the characteristic difference between these two categories. In Break 4 situation, when the lexical stress of the pitch accent wordis on the first syllable, the peak goes to the highest place than the peak of any other classified branches. Monosyllable words have the highest valley value. Because both the lexical stress feature and boundary tone are compressed to realize in this single syllable. If the valley goes lower, there might be little room to accomplish the high goal assigned by the rising tone. And at last, all pitch accent branches analyzed end at a similar pitch height.After seeing macro and micro pitch behavior of word, we would focus on a finer point: temporal behavior of pitch peak of pitch accent word, or called tonal alignment. It is a heated area in the research of prosody. Previous researches show that in English prenuclear high accent word the peak is preferentially aligned past the end of the stressed syllable. However, the materials we have found suggest in natu...
|
PDF Full Text Request