Speaker Similarities in Human Perception and Their Spectral Properties

The acoustic properties of voiced sonorants are said to be speaker-dependent, and it is also reported that these sounds are effective for identifying the speakers. In our previous experiments, we found that the nasal sounds are highly effective for speaker identification by listening, and that the inter-speaker distances in the spectra were also greater in nasals than in oral sounds. This present study further analyses the spectral properties of nasals and orals in terms of perceptual and acoustical voice similarity and shows that nasal sounds may have longer interval that listeners can exploit for speaker identification.