基于统计的语音传输指数预测方法实验对比*

良好的言语可懂度是语言声厅堂声环境设计的重要目标,语音传输指数(STI)是言语可懂度的客观评价参量。在设计阶段就能对STI做出准确预测,对语言声厅堂的声环境控制具有重要意义。基于统计的STI预测方法是IEC 60268-16标准推荐的主要预测方法,但使用该方法时的影响因素以及预测精度,当前仍缺少系统性的实验对比研究。本文使用4个房间中13个测点共52种听音条件下的STI实测结果,分析了基于统计的STI预测方法的精度以及影响因素,结果表明：(1)考虑和不考虑直达声的两种预测方法,都存在着系统偏差,预测结果普遍小于实测值；在距离声源较近测点的预测都产生了较大误差,在距离声源较远测点的预测误差相对较小；(2)基于单指数衰变声场不考虑直达声的预测方法操作简单,但误差较大,52种听音条件的平均误差达到了-0.032,最大误差达到了-0.131；(3)考虑直达声的预测方法误差相对较小,52种听音条件的STI平均误差为-0.018,最大误差为-0.080；(4)预测精度主要受所使用的信噪比、扬声器的指向性、房间的声场条件等因素影响,并对这些因素的影响进行了对比分析。结果可为STI预测方法的使用者提供参考。

Experimental comparison of speech transmission index prediction methods based on statistical

Good speech intelligibility is an important goal of acoustic environment design in lecture rooms and auditoria. The speech transmission index (STI) is an objective evaluation metric for speech intelligibility, accurate prediction of STI during the design stage is very important for the acoustic environment control in lecture rooms and auditoria. The statistical-based STI prediction method is recommended by the IEC 60268-16 standard, there is still a lack of systematic research on the influencing factors and prediction accuracy of these methods. In this paper, the influencing factors and prediction accuracy of these methods were analyzed using STI measured results of 52 listening conditions at 13 receiver positions in 4 rooms. The results revealed the followings：(1) There are systematic biases in both prediction methods with and without consideration of direct sound, and the predicted results are generally smaller than the measured results. The errors are larger when the receiver positions closer to the sound source, and relatively smaller when the receiver positions are further away from the sound source. (2) The prediction method based on single exponential decay sound field without consideration of direct sound is simple to operate, but with a larger error; the average STI error for the 52 listening conditions reached -0.032 and the maximum error reached -0.131. (3) The prediction method with consideration of direct sound has relatively small errors; the average STI error for the 52 listening conditions was -0.018 and the maximum error was -0.080. (4) The prediction accuracy is mainly influenced by the signal-to-noise ratio used, the directivity of the loudspeaker, the sound field conditions of the room, and the possible influence of these factors was compared and analyzed. The results provide a reference for the users of STI prediction methods.