Prosodic Analysis and Synthesis - Application in Computer-Assisted Language Learning

This thesis presents algorithms for automatic prosodic annotation and implementation of the analysis-by-synthesis concept of prosodic information. Therefore, a hybrid algorithm for pitch marking was developed by combining the outputs of two different pitch marking algorithms. A heuristic algorithm for pause annotation was designed by the detection of low-energy regions in speech signals. An algorithm for accent and phrase annotation was developed by decomposing pitch contours into accent and phrase commands with the Fujisaki model. The concept of analysis-by-synthesis for prosodic information was implemented by resynthesizing accents and phrases with a cepstral vocoder. Additionally, a computer-based language learning system for German learners of Mandarin was developed. A contrastive analysis of prosodic aspects of Mandarin tones between German learners and native speakers of Mandarin was performed. The effect of tone coarticulation in Mandarin on the amplitude and duration of tone commands of the Fujisaki model was examined. An HMM-based tone recognition system was redesigned by testing different spectral and prosodic features. The real-time evaluation of Mandarin tone recognizers in the language learning system was investigated.