Acoustic analysis and synthesis of nasal codas in Mandarin Chinese

The aim of this study is to identify acoustic features of the two nasal codas, [n] and [U], that exist in Mandarin Chinese. Special attention was placed on the synthesis of nasal syllables with differing numbers of pole-zero pairs. First, appropriately selected nasal utterances were copy-synthesized from the spoken utterances of a native speaker of the language. The nasal utterances were copy-synthesized with zero, one, and two nasal pole-zero pairs. A perception test was administered to provide a measure of the importance of nasal pole-zero pairs in the perception of nasals in Mandarin Chinese. Upon analysis of the collected data, it was concluded that each additional pole-zero pair improved both the perceptual quality and the accuracy of identification for the two nasal codas. Also, the high-frequency pole-zero pair provided more increase in perceptual quality in the velar nasal than for the alveolar nasal, providing a possible cue for the velar/alveolar distinction. These results should be taken with the understanding that there were potentially significant sources of error in this study. Characterizing the perceptual distinction between zero, one, or two nasal pole-zero pairs in Mandarin nasal codas will help improve the effectiveness of Chinese speech recognition and synthesis systems. Thesis Supervisor: Kenneth N. Stevens Title: Professor, Research Laboratory for Electronics