中文單詞之韻律模式研究

A Study on Prosodic Modeling for Isolated Mandarin Words

在本文中，我們對中文單詞提出了以音節為基本單位的基頻軌跡及音節長度的韻律模型。在基頻軌跡模型中，我們考慮了聲調、音節在詞的位置以及前後音節連音的三種影響因素，並假設這些影響因素彼此獨立相加而組成音節基頻軌跡。在音節長度模型中，我們考慮了聲調、音節在詞的位置、基本音節以及前後音節連音的四種影響因素，我們同樣假設這些影響因素為彼此獨立且具加成性。我們使用一個含107,936個詞的單一女性語者的語料庫來評估所提方法是否有效，我們並用決策樹來分析音節長度如何受音節音素結構的影響，也用決策樹來分析音節間pause的長度和前後音節音素結構的關係，實驗結果顯示訓練後此兩韻律模型的影響因素都符合我們對中文韻律的知識。

In this paper, syllable-based prosody modelings of pitch contour and syllable duration for isolated Mandarin words are proposed. In the syllable pitch contour model, three main affecting factors of tone, syllable position in word, and inter-syllable coarticulation are considered. These three affecting factors are assumed to be independent and additive. Similarly, in the syllable duration model, four affecting factors of tone, syllable position in word, base-syllable, and inter-syllable coarticulation are considered. We also assume that these affecting factors are independent and additive. A large single female-speaker speech database containing 107,936 words was used to evaluate the performance of the proposed methods. After well-training, the decision tree method was used to analyze the 411 affecting factors of base-syllable and to explore the relationship between inter-syllable pause duration and the nearby linguistic features. Experimental results showed that all these affecting factors conformed to our knowledge about Mandarin prosody.