2013-CIAP-Gaudrain_Fuller.pdf (3.51 MB)
Download file

Little Red Riding Hood was a Cochlear Implant User!!

Download (0 kB)
poster
posted on 07.12.2013, 23:38 authored by Etienne GaudrainEtienne Gaudrain, Christina Fuller, Jeanne N. Clarke, John J. Galvin, Qian-Jie Fu, Rolien Free, Deniz Başkent

Poster presented at CIAP 2013, Lake Tahoe, CA:

 

To track a particular speaker in a crowd, or to simply recognize someone they know (perhaps a sickly Grandma), normal-hearing (NH) listeners rely on the perception of the vocal characteristics of the speaker. Two properties seem to be particularly important: the glottal-pulse rate (GPR), which relates to pitch, and the vocal-tract length (VTL), which relates to the perceived size of the speaker. NH listeners are very sensitive to both properties and interpret GPR and VTL cues to identify speakers’ gender, age, size or other features [e.g. Smith and Patterson, 2005, JASA]. As such, how could Little Red Riding Hood (LRRH) have been fooled by the Wolf’s voice, as the VTL of the Wolf would have been much larger than that of Grandma – big enough to “swallow her whole.” We performed a series of experiments that may shed light on these long-ago tragic events, as well as on perception of speaker characteristics by cochlear implant (CI) users.

Little is known about CI users’ perception of vocal characteristics. Previous studies suggested that, for voice gender identification, CI users rely almost exclusively on GPR, while NH listeners also use VTL [Fu et al., 2004, JARO; 2005, JASA]. However, in these studies, GPR and VTL were not manipulated parametrically but estimated from recordings of actual male and female talkers. In our first experiment, we systematically manipulated GPR and VTL from a single speaker in order to eliminate all other potential cues. Voice gender identification was measured for the various GPR-VTL combinations in CI users and in NH subjects listening to non-vocoded or vocoded speech (eight-channel CI simulation). GPR was the dominant cue for CI users (real or simulated) for voice gender identification, with little contribution of VTL.

In the first experiment, it was unclear whether CI users cannot perceive VTL cues, or whether they cannot use the cues correctly to interpret the size or gender of a speaker. In our second experiment, we measured just-noticeable-differences for VTL and GPR in real CI users, and in NH subjects listening to non-vocoded or vocoded speech. Preliminary results suggest that the VTL differences typically found between males and females are not salient enough to be perceived through CIs.

If LRRH was NH, we would have to conclude that the Wolf was not as big as presumed, and did not swallow her and the Grandma “whole” (sadly preventing any happy-ending to happen). However, if LRRH was wearing a CI, the present results indicate that she would not have perceived the longer VTL of the Wolf, and would have only noticed “what a deep voice you have” (pitch) and finally “what a big mouth you have” (visual cue) right before she was eaten. Improving perception of VTL cues, maybe through cognitive training, may improve CI users’ speaker identification (and could save them from being eaten by wolves).

 

Funding: VIDI grant from ZonMw, NWO, Rosalind Franklin Fellowship, Heinsius Houbolt Foundation.

History

Usage metrics

Licence

Exports