【摘要】：本研究基于大規(guī)模語(yǔ)料庫(kù),通過(guò)剝離聲調(diào)等因素對(duì)音高的影響,逐步揭露出韻律結(jié)構(gòu)因素對(duì)漢語(yǔ)普通話陳述句音高變化的作用。并在研究結(jié)論的基礎(chǔ)上,通過(guò)對(duì)大規(guī)模語(yǔ)料庫(kù)進(jìn)行相關(guān)參數(shù)的統(tǒng)計(jì)建模,使用模型對(duì)語(yǔ)句音高進(jìn)行預(yù)測(cè),并將預(yù)測(cè)結(jié)果應(yīng)用到合成語(yǔ)音,通過(guò)MOS評(píng)測(cè)來(lái)驗(yàn)證研究結(jié)論的正確性。本文的研究?jī)?nèi)容主要包括語(yǔ)調(diào)和字調(diào)兩個(gè)方面。語(yǔ)調(diào)方面,本文將韻律層級(jí)結(jié)構(gòu)和聲調(diào)音域的概念相結(jié)合,提出了音域箱及其相關(guān)概念。本文通過(guò)統(tǒng)計(jì)分析考察了音域箱的特點(diǎn),并根據(jù)分析結(jié)果建立高音線-低音線預(yù)測(cè)模型。在已知韻律結(jié)構(gòu)和重音分布的前提下,該模型可以預(yù)測(cè)普通話陳述句的語(yǔ)調(diào)走勢(shì)。研究表明,在音域父子箱中,(1)子級(jí)箱的低線具有階梯式下行性。(2)子級(jí)箱的高線具有S-U(重音-非重音)兩級(jí)性。重音級(jí)高線明顯高于非重音級(jí)高線。(3)子級(jí)箱低線的階梯式下行性和音域箱的層級(jí)嵌套性導(dǎo)致音高重置更可能發(fā)生在等級(jí)較高的韻律邊界處,并且邊界等級(jí)越高重置幅度越大。字調(diào)方面,本文詳細(xì)分析了各個(gè)調(diào)類在不同語(yǔ)流環(huán)境中的調(diào)型,以及前字調(diào)類對(duì)本調(diào)調(diào)型的影響,并根據(jù)分析結(jié)果建立調(diào)型預(yù)測(cè)模型,在已知聲調(diào)音域和音節(jié)類型的前提下,用以預(yù)測(cè)音節(jié)的調(diào)型曲線。研究表明,(1)濁音聲母和零聲母音節(jié)的調(diào)型受前音節(jié)聲調(diào)類型影響較大,清音聲母音節(jié)的調(diào)型受前音節(jié)聲調(diào)類型影響較小。(2)前音節(jié)的調(diào)型段末尾如果具有高音特征,那么本音節(jié)的基頻段起始點(diǎn)音高較高；前音節(jié)的調(diào)型段末尾如果具有低音特征,那么本音節(jié)的基頻段起始點(diǎn)音高較低。(3)輕聲的性質(zhì)與其他幾個(gè)調(diào)類不同,它的調(diào)型和調(diào)值是依賴前音節(jié)的聲調(diào)類型而存在的。最后,本文結(jié)合上述兩個(gè)預(yù)測(cè)模型,建立了陳述句音高曲線預(yù)測(cè)模型,在已知句子韻律結(jié)構(gòu)和重音分布的前提下,用以預(yù)測(cè)普通話陳述句的音高曲線。將預(yù)測(cè)結(jié)果用于語(yǔ)音合成后得到了自然的合成效果,說(shuō)明本文的研究結(jié)果是合理的。
[Abstract]:Based on the large scale corpus, this study reveals the effect of prosodic structure on the pitch of Chinese Putonghua declarative sentences by stripping the tone and other factors on the pitch. On the basis of the conclusion of the study, through the statistical modeling of the related parameters of large-scale corpus, the pitch of sentences is predicted by using the model, and the prediction results are applied to the synthetic speech, and the correctness of the research conclusions is verified by MOS evaluation. This paper mainly includes two aspects: intonation and tone. In the aspect of intonation, this paper combines the concept of prosodic hierarchy with the concept of tone range, and puts forward the range box and its related concepts. In this paper, the characteristics of the range box are investigated by statistical analysis, and the prediction model of the treble line and the bass line is established according to the result of the analysis. On the premise of known prosodic structure and stress distribution, the model can predict the intonation trend of Putonghua declarative sentences. The results show that: (1) the lower line of the sub-box has the stepwise descending property, and (2) the high line of the sub-stage box has S-U (stress unstressed) two-level property. The stress level high line is obviously higher than the unstressed level high line. (3) the step descending property of the lower line of the sub-level box and the hierarchical nesting of the range box lead to the pitch reset more likely to occur at the prosodic boundary of the higher grade. And the higher the boundary level, the larger the reset range. In terms of tone, this paper analyzes in detail the tone types of each tone class in different language flow environments, and the influence of the former tone type on the tone type. Based on the results of the analysis, a diatonic prediction model is established, which is based on the known tone range and syllable type. A phonic curve used to predict syllables. The results show that: (1) the tone types of turbid consonants and zero consonants are greatly influenced by the phonetic types of antecedent syllables, while those of the consonant syllables are less influenced by the phonetic types of presyllable syllables. (2) if there is a high tone characteristic at the end of the tone segments of the phonetic syllables, The starting point of the base band of this syllable is high pitch; If there is a low pitch at the end of the tone segment of the pre-syllable, the initial pitch of the base band of this syllable is low. (3) the nature of the soft tone is different from that of other tone classes, and its tone type and modulation value are dependent on the tone type of the prosyllabic. Finally, combining the above two prediction models, this paper establishes a pitch curve prediction model of declarative sentences, which can be used to predict the pitch curves of declarative sentences in Putonghua on the premise of known prosodic structure and stress distribution. After applying the prediction results to speech synthesis, the natural synthesis effect is obtained, which shows that the research results in this paper are reasonable.
【學(xué)位授予單位】：復(fù)旦大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2013
【分類號(hào)】：H116

【參考文獻(xiàn)】

相關(guān)期刊論文 前2條

1 黃賢軍;高路;楊玉芳;呂士楠;;漢語(yǔ)語(yǔ)調(diào)音高下傾的實(shí)驗(yàn)研究[J];聲學(xué)學(xué)報(bào)(中文版);2009年02期

2 沈炯;漢語(yǔ)音高系統(tǒng)的有聲性和區(qū)別性[J];語(yǔ)言文字應(yīng)用;1995年02期

，

本文編號(hào)：2385972