Acoustic parameters in human speaker recognition.

Abstract

Four speaker identification tests were conducted using five female speakers known to the listeners. Starting from acoustic recordings of reiterant "ma" syllables, the perceptual importance of the following three factors was investigated: F0 height, F0 contour, and speech rhythm. For speakers with typically low or high voices F0 height turned out to be a highly relevant cue in speaker identification. For all speakers F0 contour was of secondary importance, whereas speech rhythm had a small but consistent influence on recognition rates. It could be inferred that remaining factors alone (mainly global spectral information) would yield recognition scores of approximately 50%. Consistent with previous investigations, the relevance of perceptual cues in the recognition of familiar voices was shown to be not hierarchically fixed, but to depend on speaker-specific voice characteristics.