| Popis: |
This study is about an approach that formulates an explicit way from continuous acoustic parameters to discrete and abstract phonological entities. The method is implemented in a computer program and uses a linguistic theory about the underlying structure of prosody in speech. The program is designed to automatically detect the position of prosodic events from acoustic speech signals. Such a program can be of great benefit for the linguist working with large acoustic databases. It enables the researcher to process unlabeled speech material automatically and systematically. The program can search for specificintonational patterns in a given language, or can test a theory about the underlying structure of prosody against the acoustic reality or the language learner can use it by seeing some visual feedback to his or her freshly acquired foreign language abilities. Furthermore the program can be used for labelingprosodic events in a spoken speech synthesis corpus and consequently improve the synthesis quality. Last but not least there are possible applications in the field of automatic speech recognition.The process of information extraction from acoustic speech signals involves not only the recognition of segmental features, phonemes, syllables or words and subsequent linguistic processing, but also the recognition of prosodic events including the position of accented words, the type of pitch movement associated with them, the general trendline of pitch and also the grouping of information units, phrases or words.The prosodic events are important conveyors of the information structure in utterances, which this work aims at unfolding for improved speech analysis and recognition. To fulfill these aims, the following tasks are done: (i) review and discussion of intonation models, (ii) development of a new approach for the automatic detection of prosodic cues, (iii) acoustic analysis of cues of prosodic events, (iv) implementation of algorithms for detecting theseprosodic cues, and (v) evaluation of the new approach. Important aspects of the thesis include integration and evaluation of linguistic theory and quantitativeacoustic modeling. |
