強勢アクセント・高さアクセント・音調（それぞれ、英語・日本語・中国語など）といった語彙アクセントの類型や、強勢リズム・音節リズム・拍リズム（それぞれ、英語・スペイン語・日本語など）といったリズムの類型は、よく知られた概念であるが、それが音響的に定義できるかどうかは分かっていない。本稿では、音源フィルタモデルの音源部（音響的特徴）が韻律（言語学的特徴）にほぼ対応するという仮説を、実験音声学的な研究によって論じる。
本研究は４つの部分から成る。まず、予備実験で、英語と日本語のスペクトル情報（非音源部）を削減した音声で言語の識別ができることを示す。次に、音源部を用いた子音の聴取実験により、リズムにとって重要な主要音類の情報が音源部に含まれることを示す。そして、韻律類型が異なる中国語・英語・日本語・スペイン語の音響分析により、音源部が異なっていることを示す。最後に、これら４言語の音源部をパラメタ化した聴取実験で、音源部の情報によって４言語が識別できることを示す。
これらの結果は、音源部と韻律の対応関係を示している。本研究は、通言語的な韻律の記述システム構築のための基礎となることを目的としている。
Different languages sound different, and considerable part of it derives from the typological difference of prosody. Although such difference is often referred to as lexical accent types (stress accent, pitch accent, and tone; e.g. English, Japanese, and Chinese respectively) and rhythm types (stress-, syllable-, and mora-timed rhythms; e.g. English, Spanish, and Japanese respectively), it is unclear whether these types are determined in terms of acoustic properties. The thesis intends to provide a potential basis for the description of prosody in terms of acoustics. It argues for the hypothesis that the source component of the source-filter model (acoustic features) approximately corresponds to prosody (linguistic features) through several experimental-phonetic studies.
The study consists of four parts. (1) Preliminary experiment: Perceptual language identification tests were performed using English and Japanese speech samples whose frequency spectral information (i.e. non-source component) is heavily reduced. The results indicated that humans can discriminate languages with such signals. (2) Discussion on the linguistic information that the source component contains: This part constitutes the foundation of the argument of the thesis. Perception tests of consonants with the source signal indicated that the source component carries the information on broad categories of phonemes that contributes to the creation of rhythm. (3) Acoustic analysis: The speech samples of Chinese, English, Japanese, and Spanish, differing in prosodic types, were analyzed. These languages showed difference in acoustic characteristics of the source component. (4) Perceptual experiment: A language identification test for the above four languages was performed using the source signal with its acoustic features parameterized. It revealed that humans can discriminate prosodic types solely with the source features and that the discrimination is easier as acoustic information increases.
The series of studies showed the correspondence of the source component to prosodic features. In linguistics, prosodic types have not been discussed purely in terms of acoustics; they are usually related to the function of prosody or phonological units such as phonemes. The present thesis focuses on acoustics and makes a contribution to establishing the crosslinguistic description system of prosody.