The effect of voicing, pitch, and vocal tract length on the recognition of concurrent speech

Abstract

In natural speech, the pitch of the voice can be varied by the speaker, but they cannot vary their vocal tract length (VTL), so VTL would seem to be a logical choice of tracking variable for listening in multispeaker environments. Experiments with concurrent speech have shown that differences in pitch and VTL make it possible for listeners to attend to a particular speaker in a multi-speaker environment, but the relative contribution of pitch and VTL is not clear. A series of experiments was conducted to investigate the extent to which differences in vocal characteristics contribute to the recognition of concurrent speech, using syllables with matched temporal envelopes to preclude glimpsing. Syllable pairs were synthesized with a vocoder to simulate different speakers. We measured the recognition of whispered and voiced syllables masked by syllables that varied in voicing, pitch, and/or VTL. The results show that differences in any of the vocal characteristics improve recognition performance. Voiced speech was more recognizable than whispered speech at the same SNR when it was the target speaker. When the distracter was whispered speech, it produced lower recognition scores than voiced speech. [Research supported by the UK-MRC (G0500221, G9900369).]