A High-accuracy Approach to Pronunciation Prediction for Out-of-vocabulary English Word

摘要

Letter-to-Sound conversion is one of the fundamental issues in text-to-speech synthesis. In this paper, we address an approach to automatic prediction of word pronunciation. This approach combines example-based learning and dynamic-programming searching to predict sub-word pronunciation. Word pronunciation is formed by concatenating sub-word pronunciations. We conducted comparative experiments over a large-scale English dictionary. Experimental results show that this approach can achieve accuracy of 70.1%, which outperforms those published results.