Study of perceptual factors for speaker identification focusing on perceptual similarity of speaker characteristics

摘要

The perceptual factors contributing to speaker identification, focusing on perceptual similarity for spoken sentences were investigated. Five male and five female listeners participated in the experiment. They had never met the speakers or listened to the speakers' voices. All were native speakers of Japanese and had no known hearing impairments. The participants were presented with a pair of sentences uttered by two of the speakers. The participants were asked to evaluate the similarity between the speakers of each pair of sentences on a four-level scale of 'similar,' 'rather similar,' 'not very similar,' and 'dissimilar.' On the basis of the participants' comments, the pitch frequency contour of the speech data between the speakers was compared. The relative contribution of each factor depended on the pair of speakers, suggesting that the acoustic features that differ considerably between two speakers in the perceptual space contribute to speaker identification.