Applying Hybrid Genetic Algorithm in Solving the Polysemy Problem in a Chinese to Taiwanese TTS System

• Main Authors: Chia-Ming Chang, 張家銘
• Other Authors: Ming-Shing Yu
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/12586170568860065621

碩士 === 國立中興大學 === 資訊科學與工程學系所 === 99 === A novel method is proposed for solving the polysemy problem in Taiwanese Text to Speech (TTS) System based on unsupervised learning. In Taiwanese, one word possibly has several pronunciations. It leads to that Taiwanese TTS System can''t work well if wrong pronunciation is synthesized. Thus, we need to decide which pronunciation is proper under some condition in order to enhance the performance of a Taiwanese TTS System. In our previous work, we use supervised learning to predict Chinese pronunciation. Experimental results show that the approaches work well. However, the polysemy problem in Taiwanese is more complex than that in Chinese . It is very time comusing in labeling the proper pronunciation for each experiment data. Such that it is not practical in implementing a Taiwanese TTS system. In order to improve the drawback mentioned above, we propose a method of combining E-HowNet and the Hybrid Genetic algorithm to solve the polysemy problem in a Taiwanese TTS system. There are six words applied in proposed approach in this paper. They are『你』(you),『我』(I),『他』(he),『上』(up),『下』(down),『不』(no). The accuracies are 93.90%, 95.28%, 96.25%, 87.05%, 86.56%, and 62.51%, respectively. The experimental results show that the proposed approach can reduce at least 98% of the work in labeling the corpus.