基于汉语语音素材的响度算法误差研究

Study of deviations of different loudness algorithms based on Chinese speech materials

对两类在国际上得到广泛接受的响度模型即Moore和Zwicker时变响度模型及ITU(国际电信联盟)标准推荐响度模型进行了对比研究.通过耳机回放的主观听音实验评估了这两类响度模型在计算汉语语音素材总响度时的有效性.模型的评价采用了4种不同的统计量,对模型的绝对精度和相对精度进行了度量.研究发现,ITU-R BS.1770-4推荐响度模型相对简单,更易于对语音素材的总响度进行度量;在计算汉语语音素材的总响度时,ITU标准推荐响度模型比Moore和Zwicker时变响度模型的精度高,提高了0.5～2 dB.

A comparative study has been completed between two types of internationally recognized loudness models: Moore′s and Zwicker′s time-varying loudness models and loudness models recommended by ITU(International Telecommunication Union). To test the availability of these loudness models in terms of calculating the total loudness of Chinese speech materials, a loudness listening experiment is realized, and acoustic stimuli are presented to subjects through headphones. Four different statistical measures are employed in the evaluation of the models, and the absolute accuracy and relative accuracy of the models are measured. It is found that the loudness model recommended by ITU-R BS. 1770-4 is relatively simple and easier to measure the total loudness of speech materials. When calculating the total loudness of Chinese speech materials, the loudness models recommended by the ITU are more accurate than Moore′s and Zwicker′s time-varying loudness models, and the improved accuracy is 0.5-2 dB.