基于语音起始段检测语音可懂度客观评价方法

Objective measures for predicting speech intelligibility in noisy conditions based on speech onset detection

传统的语音评价算法,如SNR,存在语音的可懂度相关性不高的问题。有研究表明,语音的不同部分对可懂度的贡献不同,语音的浊音起始段对可懂度的影响较大。提出一种可懂度相关性相对较高的语音评价算法。在计算分段SNR之前,对语音段进行选择,选出起始段。所提出方法的可懂度计算结果与主观得分进行比较,实验结果表明,结合语音起始段(speech onset)检测算法,能够将可懂度与主观评价的相关值分别提高0.11(辅音)和0.06(句子),这也从一个侧面验证了语音的起始段对可懂度有较大影响这一研究结论。

Traditional speech objective measure, like SNR, has a poor correlation with speech intelligibility. Studies have shown that the different parts of speech have different contributions. Speech onset has a greater impact on speech intelligibility. Think about it, this paper presents a speech objective measure which has a relatively high correlation with speech intelligibility. Be- fore computing segment SNR, speech and the relative onset shouid be selected correctly and precisely. The proposed measures were evaluated with intelligibility scores obtained by normal-hearing listeners.The results from this study clearly suggest that correlations between speech intelligibility and subjective evaluation could relatively increased by 0.11 （consonant） and 0.06 （sentence） with the proposed speech onset detection. It verifies the study result that speech onset does have a great impact on speech intelligibility.