Collecting Mandarin Speech Databases for Prosody Investigations.
Chiu-yu Tseng · Journal of Chinese Language and Computing · 2004
The prosody of Mandarin running speech is notably marked by grouping of short phrases into perceptually identifiable larger units in the speech flow. An organization of Mandarin speech prosody should not only account for the grouping phenomenon, but also offer some explanation for such grouping in relation to information of other linguistic levels as well as speech planning. The physical, phonetic, acoustic, semantic and syntactic characteristics prosodic units as well as their perceptual properties have been under investigation at our lab. How these units may relate to and combine with preceding as well as following silent portions in the speech flow to constitute the overall phrasing of running speech in general, and how they could be viewed from the perceptive of speech planning in particular have also been the focus of our investigation. We are very much aware of the fact that prosody varies for different speech styles, and therefore devised methods to collect speech corpora of different speech styles. Since our aim was to capture the characteristics that constitute the overall flow and rhythmic structure of connected speech, our speech samples were long utterances of prosodic phrases, utterances and prosody groups in different durations, and were longer than utterances normally found in syntactic investigations. In this paper, we will report methods we devised to collect speech data of the following speaking styles: read speech by untrained native speakers, read speech by radio announcers, spontaneous speech of specific topics and without topic specification, spontaneous speech of public speaking and monologue. Speech data included microphone speech recorded in sound proof chambers and telephone speech. Text used included well structured paragraphs as well as word salads.