朝鲜语语音音节自动切分算法的研究

针对目前语音语料人工标注效率低的问题,提出了一种朝鲜语连续语音语料的音节自动切分方法.该方法首先采用Seneff听觉模型提取音频的包络检测响应和广义同步检测响应等特征参数,其次结合朝鲜语发音特点确定音节的候选边界位置,最后通过静音段和摩擦音检测消除虚假边界,以提高边界检测的准确率.实验结果表明,该朝鲜语语音语料音节自动切分方法的准确率(93.56%)比传统的基于Seneff听觉模型的分割算法提高了14.59%,召回率(86.43%)比传统的基于Seneff听觉模型的分割算法降低了1.69%; 因此,本文算法总体优于传统的基于Seneff听觉模型的分割算法.

Research on automatic segmentation algorithm of Korean speech syllables

Aiming at the current low efficiency of manual annotation of speech corpus, an automatic syllable segmentation method for Korean continuous speech corpus is proposed. First, Seneff auditory model is used to extract the audio characteristic parameters, such as the envelope detection response and generalized synchronous detection response, etc. Secondly, the candidate boundary position of syllables is defined according to the Korean pronunciation characteristics. Finally, the false boundary is eliminated by silent segment and fricative detection to improve the boundary detection accuracy. The experimental results show that the accuracy of the proposed Korean syllable segmentation method is 93.56%, increased by 14.59% than that of traditional segmentation algorithms based on Sneff auditory model, meanwhile, the recall rate reaches to 86.43%, decreased by 1.69%. Therefore, the proposed algorithm in this paper is overall superior to traditional segmentation algorithms based on Sneff auditory model.