Abstract: Speech intelligibility is an important attribute of speech signal. Based on Normalization Covariance Metric（NCM）, the speech signal is segmented with a relative Root Mean Square（RMS） threshold, The segmentation intelligibility evaluation is performed for speech segments above the mean square value and for speech segments below the mean square value. At the same time, this paper presents a new intelligibility evaluation model, which combines the relative contribution of the two speech segments to the intelligibility of speech, and evaluates the intelligibility of speech. The experimental results show that the high-mean speech segment has a higher contribution to the intelligibility than the low-mean square speech segment, and the evaluation results of the two speech segments are re-combined with the new model, and the evaluation effect is improved significantly.

Key words: speech intelligibility, segmentation intelligibility evaluation, relative root mean square, evaluation model

摘要： 语音可懂度是语音信号的一种重要属性，在归一化协方差评价方法（NCM）的基础之上，以相对均方根（RMS）为阈值对语音信号进行分割，对高于均方值的语音段和低于均方值的语音段进行了分段可懂度评估，同时，提出了一种新的可懂度评价模型，结合了这两种语音段对语音可懂度的相对贡献，共同评价语音的可懂度。实验结果表明，高均方语音段相对于低均方语音段对可懂度具有更高的贡献，利用新的模型将这两种语音段的评价结果进行重新结合，评价效果得到了显著提升。

关键词: 语音可懂度, 分段可懂度评估, 相对均方根, 评价模型