【導讀】為英文輸入生成詞組水平音高和時間持續(xù)久的信息。這一信息是用來注釋輸入句。子，然后被處理的文本到語音程序目前在貝爾實驗室開發(fā)。們的目標一直是檢驗假設(shè)該語法樹中的信息可用。尤其地，如主謂和頭補。能它可以使用一個指定語法句法分析函數(shù)來確定合成語音的韻律短語。雖然語法和韻律之間的某些關(guān)聯(lián)是眾所周知的（例如像進度話詞性應(yīng)力的影響，在許多研究中，研究人員之間尋求成分結(jié)構(gòu)和韻律連接（如Cooper. Gee和Grosjean1983）但是，隨著。他們往往忽略了在svntax樹語法功能的代表性。此外，以前的工作還沒有具體明確，提供了一個完整的系統(tǒng)實施的基礎(chǔ)。已實施了韻律規(guī)則的集合在一個實驗文語轉(zhuǎn)換系統(tǒng)。音的生產(chǎn)是可行的。但沒有試圖代表附屬或修飾關(guān)系。其結(jié)果是一個支持該觀點的分層表。律中能夠產(chǎn)生顯著改善。嚴重違反我們的基本方針。構(gòu)，從句法樹決定韻律邊界，但聲明不是明確提出了一個句法成分。

　　

【正文】 the longer the stored segments selected for the concatenation, the more natural the generated speech. As each speech unit may have many variants in different contexts or prosodic situations, this approach needs a large memory to store a large number of speech segments. The approach was not practical some years ago because of the limitation of puter power and memory. With the development of hardware, the use of large speech corpus as synthetic units for direct concatenation is possible. Unit selectionbased speech synthesis (or corpusbased synthesis) has been applied in English and other languages for some years. Some attempts (Liu, and Wang, 1998。 Chu et al. 2020。 Wang et al., 2020, Li et al, 2020) have been made for Chinese TTS using unit selection approach. Wu et al. (2020) also proposed a scheme to select phoically, linguistically best units and then apply prosodic modifications. However, all the proposed approaches have limitations in the application of proper prosody. Without proper prosody consideration, the quality of the generated speech may be poor sometimes. This paper concerns about how to apply prosody in a unit selection based synthesis. 2 Unit Selection Model A unit selection model has a wellorganized unit database. The database contains the speech units from a large corpus, which is carefully designed to have a large coverage of all phoic and prosodic variants of each unit. In the database, each speech unit has a number of possible variants, which are suitable to appear in different phoic and prosodic environments. The large speech corpus is analyzed offline and all the calculated features are stored in a unit database. In the database, each instance of a unit is described by a vector of features. Each feature may be a discrete or continuous value. The features include features of the unit itself and the context of the unit. The features of the unit itself are used for selecting the correct unit that meets the segmental requirement, while the features of context are used for selecting the contextually best unit, which may minimize the discontinuity between the selected units. The corpusbased concatenation synthesis is actually a pattern matching process. During synthesis, the work need to do is to select the best units that phoically and prosodically best match the target units. Meanwhile, the discontinuity between selected units should be kept as small as possible. To meet these requirements, two costs should be defined in synthesis. One is unit cost, which describes how close a selected unit to the desired unit. The other is connection cost, which describes the degree of continuity between the selected units. The whole cost is the weighted sum of the two costs. 3 Unit Selection Process The speech synthesis process accepts information from prosody generation part, retrieves the speech unit database to find a proper unit for every target speech unit. The unit selection process can be illustrated as Figure 1. In the figure, the target sentence is “今天很熱 (it is very hot today)”, which consists of 4 syllables. Each syllable has a set of candidate units. The thick line and thick edge box indicate the selected unit sequence. In unit selection process, to get the best speech, we have to consider (1) the properness of the candidate unit pared with target unit, (2) the smoothness between the selected units to be connected. Therefore, the selection process is to find a best path among all the possible paths in the connection lattice. The search process is guided by a cost function, which describes the degree of properness of a unit and degree of smoothness between two units. 4 Corpus As we have mentioned earlier, a large speech corpus is used in unit selection based synthesis. The speech corpus consists of a large collection of utterances. The unit for the synthesis will be extracted from the corpus. It is ideal to cover context dependent units and prosody variants as much as possible. However, it is usually impossible to build a very large speech corpus that has a plete coverage of unit variants. As the cost of constructing a large corpus with high quality is very expensive, balance is usually made between coverage and size. In this research, we built a corpus of around 38000 syllables. The script of this speech corpus is selected from a large text corpus (around 300M Chinese characters). The corpus is designed to cover the frequently used context independent syllable and context dependent syllable as much as possible. We use PKU People’s Daily text corpus as a reference for real word text to evaluate the script of the corpus. We calculated that the built corpus covers % of syllable occurrences in the PKU corpus. When context of unit is grouped by initial and final class (we defined 11 Initial classes and 10 final classes), the speech corpus covers % of the classes of unit occurrences in PKU text corpus. With such coverage, we think the corpus is proper for unit selection based synthesis.