摘要
在英文TTS(texttospeech)系统中 ,需要根据文本中每一个单词的发音来合成语音 由于在真实文本的处理中 ,无论词典规模如何大 ,都不可能包括文本中的每一个单词 ,所以需要使用某种算法来预测词典中未登录单词的发音 介绍了一种基于实例学习的方法 ,并在一个大规模的英语词典上进行了性能评测 结果表明 ,这种方法的单词发音正确率可以达到 70 1% 。
In TTS(text to speech)systems, the pronunciation of each word is needed to synthesize the voice Because every word in the text can not be listed exhaustively when processing the real world documents, no matter what the scope of dictionary is, some kinds of algorithms are needed to automatically predict the pronunciation of word which is not included in the lexicon In this paper an approach based on exemplar learning is introduced and its performance evaluated on a large scale English dictionary Experimental results show that this method can achieve accuracy of 70 1%, obviously higher than the published approaches
出处
《计算机研究与发展》
EI
CSCD
北大核心
2004年第5期796-801,共6页
Journal of Computer Research and Development
关键词
机器学习
实例学习
machine learning
exemplar learning