语音中相位的听觉感知实验研究
Experimental study on phase perception in speech
-
摘要: 人的听觉对语音信号中相位的感知比较迟钝,因而对语音信号进行处理和编码时常常不关心相位失真。实际上,相位失真到一定程度时会明显导致语音质量的下降。为了取得高质量的声码器,语音谱分量的相位信息是不能不考虑的。本文通过主观听觉测试实验研究了语音信号的短时Fourier变换相位谱对人的听觉感知的影响。测试结果表明:(1)如果完全舍弃原相位信息,则得到的重建语音含有很强的噪声且自然度很差;(2)不论舍弃高频段还是低频段的相位信息,均能导致听觉感知差异;(3)当相位的量阶小于π/7时,人的听觉系统将分辨不出重建语音和原始语音之间存在的差异.Abstract: As the human ear is dull to the phase in speech, little attention has been paid to phase information in speech coding. In fact, the speech perceptual quality may be degrated if the phase distortion is very large. The perceptual efiect of the STFT phase spectrum is studied by auditory subjective tests in this paper. Three main conclusions are: (1) If the phase information is neglected completely, the naturalness of the reconstructed speech is very poor; (2) Whether the neglected phase is in low frequency band or high frequency band, the difference from the original speech can be perceived by ear; (3) The human ear can not perceive the difference of speech quality between original speech and reconstructed speech while the phase quantization step size is little than π/7.