Abstract
In this study, we have implemented a word-segmentation system for text-to-speech synthesis in Mandarin Chinese. The proposed word-segmentation system consists of several functional blocks that can be used to decompose a given sentence into natural "word units" for achieving natural prosody in text-to-speech synthesis. We propose two methods for basic word segmentation, including maximum matching algorithm based on longest word first, and dynamic programming based on word length and frequencies.Moreover, we have applied other schemes to further improve the system's performance, including word construction based on determinative measure, reiterative patterns, and name entities. We also tried to expand the word database via mutual information of on-line news articles. The tone variations that are specific to Mandarin Chinese are also addressed.