论文部分内容阅读
重音是语音合成中韵律处理的一个重要参数。本文分析了轻声和重读音节同正常重音在各声学参数上的差异,包括基频、音节时长、强度、停顿长度等,还特别考察了时长同基频参数之间的关系,以及上声音调同基频的关系。建立了基于人工神经网络的三种重音预测模型,即声学预测模型、语言学预测模型和混合预测模型,对汉语句重音(包括轻声、正常重音、重读)进行了自动判别,结果显示混合模型要优于另外两种模型。此外,本文还根据重音标注的多样性现象设计了支持率的评价方法。
Stress is an important parameter of prosody processing in speech synthesis. In this paper, we analyze the difference of acoustical parameters between syllables and normal accents, including the fundamental frequency, syllable duration, intensity, pause length and so on. We also examine in particular the relationship between duration and fundamental parameters, The relationship between the fundamental frequency. Three kinds of stress prediction models based on artificial neural network, namely acoustic prediction model, linguistic prediction model and mixed prediction model, were established to automatically discriminate Chinese sentence stress (including soft tone, normal stress and reread). The results showed that the mixed model Better than the other two models. In addition, this article also based on the phenomenon of stress annotation design support rating evaluation method.