下載5知识课件.ppt

台大課程語料的韻律組織與特性;研究（討論）重點;;台大課程語料總覽;台大課程語料內容;介面（1/2）;介面2;Speech Samples of Spontaneous Classroom Lecture;NTU課程語料特點;語料特性初步觀察（1/3）;語料特性初步觀察（2/3）;語料特性初步觀察（3/3）;Classroom lectures 特性（Yamazaki, Iwano, Shinoda, Furui and Yokota, Interspeech2007);Yamazaki, Iwano, Shinoda, Furui and Yokota (2007).;Where and How to Look for Prosodic Patterns? ;Q: IS and Prosody;Lecture Speech—DS vs. IS Interaction between discourse coherence (DC) and information structure (IS) in SpnL ;Discourse Planning of Spontaneous Lecture vs. Read Speech;Preliminary Comparison between RS and SpnL;Speech Materials;Speech Samples;Speech data used were processed manually by trained transcribers for perceived boundaries by HPG protocol and annotation including (1) speech paragraph (2) intermediate intonation phrases All CNA, CL and WB have been annotated completely for (1) and (2). 90 minutes of NDLC was labeled for (1) and 1 hr of NDLC was labeled for (2) Speech data were annotated manually with perceived emphasis for analysis as well. So far, 1 hr of NDLC, 1 male speaker of WB and 1 female speaker of CNA were selected for emphasis annotation. ;The Speech Materials by Analysis ; Mean length per BG (complex sentence and change of breath) in by data type and speaker Mean length per PG (speech paragraph) in temporal measurement by data type and speaker A BG of SpnS could be 3 to 4.5 times longer than read speech; a PG 26 to 40 times more than read speech CNA and CL. ;Scale of Discourse Planning by Number of Syllables ;Indicator R mean number R of BG within one PG by data type and speaker In SpnS the mean number of complex sentences per paragraph is 6 while in read speech CNA and CL 2 to 3. ;Global PPh F0 Pattern Using the Fujisaki Command Ap low pass filter for automatic phrase segmentation Pause Duration & tempo by PPh Intensity Average by PPh Definition of Contrastive Feature Statistical Analysis ;Ap Comparison for Discourse Boundary Discrimination by Relative and Singular Cues;Cross-