语音合成中的语言学计算模型:现状及展望

Linguistic models in speech synthesis:The state of the art

本文对应用于语音合成的文本处理技术的现状进行了总结,其中涉及了三个计算模型:自动分词器、自动标音器、韵律预测器。在分析现阶段语音合成系统存在的与文本处理模块相关联的注音错误、韵律结构预测偏差、韵律描述过于简略等三方面问题之后,指出在语言学计算模型中必须应用更为丰富的语言学知识,并提出为解决这些问题可能的技术实现路径。

This paper examines the state-of-the-art technology in text processing modular used in speech synthesis systems.It includes three computing models:parser,phonetizer and prosodic structure predictor.Problems associated with phonetic error occurrence,prosodic structure prediction and prosodic description in current systems are analyzed.As solutions to these problems,richer linguistic information is proposed to be integrated into the linguistic models and methodologies for achieving the goals are also explored.