Estimation of environmental conditions based on acoustic signals for speech dialog systems

点击翻译按钮获取中文摘要

The conditions of users and environments are not considered much in speech dialog systems. We propose the estimation method of environmental conditions based on acoustic signals. The distance from a user to a system is estimated using estimated an acoustic transfer function. At first, templates of frequency characteristics depending on the distance are created from impulse responses for each distance. To estimate the transfer functions, we calculate the difference between input speech and clean speech using clean speech models. This difference means the transfer function depending on the distance. The distance can be estimated from the nearest template. The almost 80 % accuracy was obtained in the discriminative experiments that was to decide if the distance is closer or far than 1 m.