论文部分内容阅读
该文基于优化的检测网络和多层感知(multi-layerperception,MLP)特征,提出一种可以更加准确地检测出错误发音类型的方法。首先,从第二语言学习的语音库中提取出基本的发音规则以及组合的发音规则,并相应地计算它们发生的先验概率,再将这些具有先验概率的规则用于构建基于多发音的扩展检测网络。然后在检测过程中,引入基于发音特征的MLP特征来描述发音概率,替代了传统的语音声学特征。最后使用基于MLP特征的GMM-HMM框架从检测网络中识别出最可能的发音音素串。实验表明:该方法将音素识别正确率提高了3.11%,错误类型准确率提高了7.42%。
Based on the optimized detection network and multi-layerperception (MLP) features, this paper proposes a method that can detect the incorrect pronunciation types more accurately. First of all, the basic pronunciation rules and the combined pronunciation rules are extracted from the speech database of the second language learning, and the prior probability of them is calculated accordingly, and then the rules with prior probability are used to construct the polyphony-based Expand the detection network. Then, MLP features based on pronunciation features are introduced to describe the pronunciation probability in the detection process, which replaces the traditional speech acoustics features. Finally, the GMM-HMM framework based on MLP features is used to identify the most probable phoneme strings from the detection network. Experimental results show that this method improves the phoneme recognition accuracy by 3.11% and the error type accuracy by 7.42%.