基于听觉模型的耳语音的声韵切分

Initial/final segmentation of Chinese whispered speech based on the auditory model

本文分析了耳语音的特点,并根据生理声学及心理声学的基本理论与实验资料,提出了一种利用听觉模型来进行耳语音声韵切分的方法。这种适用于耳语音声韵切分的听觉感知模型主要分为四个层次:耳蜗对声音频率的分解机理;听觉系统的时域和频域非线性变化;中枢神经系统的侧抑制机理。这种模型能反映在噪声环境下人对低能量语音的听觉感知特性,因而适于耳语音识别,在耳语音声韵母切分实验中得到了满意的结果。

In this paper, the characteristics of whispered speech are discussed, and a new approach for initial/final segmentation of Chinese whispered speech is proposed on the basis of psychological acoustic theories and experiments. With the mainly four levels of signal processing, this model can represent human's perceptual features of low energy speech, so it is more suitable for the whispered speech recognition. With the experiments of the division between the initial and the final of whispered speech included 386 Chinese syllables at 5dB SNR, the results show that the proposed approach can catch the features of whispered speech more accurately.