7th International Conference on Spoken Language Processing
September 16-20, 2002
An experiment was carried out to investigate the contextual effects of linguistic features on segment duration in spoken Korean. Contextual effects such as the segmental effects of surrounding segments, the prosodic phrasal effects and the syllable structure were considered for the investigation. This paper concentrates on segment duration analysis of a news-reading speech style, using a corpus of 670 read sentences collected from one speaker of standard Korean. Classification and Regression Tree (CART) analysis was used to explore the relationship between the context features and the realised duration. Results showed that prosodic phrase features had significant effects on segment duration. Most of the prosodic phrase influences were attributed to the accentual phrase boundary lengthening effect. Vowels either preceding or following the [nasal] feature were significantly shortened. More shortening effects were observed in the nasals than in the homorganic voiced obstruents. The [stiff vocal fold] feature, which covers the aspirated obstruents and tense obstruents in Korean, shortened the following vowel considerably. Subsequent to these linguistic findings, duration models based on Sums-of-Products (SoP) models and CART models were built for textto- speech conversion.
Bibliographic reference. Chung, Hyunsong (2002): "Segment duration in spoken korean", In ICSLP-2002, 1105-1108.