言语可懂度简易客观评价方法对比

Comparison on Simple Objective Evaluation Methods of Speech Intelligibility

言语可懂度是厅堂音质评价的重要指标。对言语可懂度的客观评价,除国际标准推荐的语音传输指数(Speech Transmission Index,STI)外,在工程实践中,声学设计者常使用一些易于测量和掌握的简易方法,如明晰度C_(50)、清晰度D_(50)、有益有害声能比U_(50)以及语音干扰级等。结合大量教室实测数据,通过建立简易方法与STI的线性关系,系统地对比简易方法相对于STI的偏差程度,结果表明:各简易方法在不同声学条件下与STI的偏差程度存在较大差异,在某些条件下可与STI媲美;C_(50),D_(50)在混响时间较长时与STI偏差程度较小,线性回归结果的估计标准误差(Standard Error of Estimate,Se)均小于STI的1个JND;U_(50)在不同混响和信噪比条件下与STI的偏差程度均较小,线性回归得到的Se最大仅在STI的1个JND左右;语音干扰级在混响时间较短时与STI偏差程度相对较小,但Se仍接近2个JND。相关结果可为简易方法的使用者提供参考。

Speech intelligibility is an important indicator for the evaluation of the sound quality of the hall.For the objective evaluation of speech intelligibility,in addition to the Speech Transmission Index(STI)recommended by international standards,in engineering practice,acoustic designers often use some simple methods that are easy to measure and master,such as C_(50),D_(50),U_(50),speech interference level,etc.This paper combines a large number of classroom measurements and systematically compares the deviation of simple methods relative to STI by establishing a linear relationship between simple methods and STI.The results show that,the degree of deviation from STI under different acoustic conditions varies considerably among the simplified methods and is comparable to STI under certain conditions,C_(50)and D_(50)deviate less from STI at longer reverberation times,and the Standard Error of Estimate(Se)of the linear regression results are less than 1 JND of STI,the deviation of U_(50)from STI for different reverberation and signal-to-noise ratios is small,and the maximum Se of the linear regression results is only around 1 JND of STI,speech interference level deviates relatively little from STI at shorter reverberation times,but Se is still close to 2 JND.The results can provide references for users of the simple methods.