Talker identification training using simulations of acoustic and electric hearing: Generalization to speech recognition

Vidya Krull, Purdue University

Abstract

In this study, we examined the effect of adding low-frequency acoustic cues to a cochlear implant (CI) simulation on talker identification within the context of a 4-day talker identification training paradigm. Pre- and post-test measures included sentence recognition for both trained and untrained talkers in quiet and in speech babble noise, as well as emotion recognition for novel talkers. One group of normal-hearing subjects listened to an 8-channel CI simulation in their right ear, whereas another group additionally listened to 500-Hz low-pass filtered speech in the left ear, simulating a CI combined with a hearing aid (HA) in the contralateral ear (CI+HA). The CI+HA group showed more improvement in talker identification accuracy with training than the CI group, even when the 500 Hz low-pass speech was replaced by an F0-controlled sine wave. However, F0 cues alone only partially accounted for the benefits. For both groups, sentence recognition for trained and untrained talkers improved with training, both in quiet and in noise. Compared to the CI group, the CI+HA group showed better sentence recognition before and after training, but their improvements in quiet were smaller compared to the CI group, likely due to ceiling effects. However, in noise both groups showed similar improvements. These improvements were retained in both groups a month post-training. The CI+HA group performed significantly better than the CI group on the emotion recognition task before and after training and overall post-test performance was significantly better than pre-test performance. These results suggest that training may improve talker identification for CI users with or without access to low-frequency acoustic hearing. However, training was more effective with access to low-frequency acoustic hearing, in terms of improvement in talker identification and generalization to both speech and emotion recognition. These findings have significant clinical implications for CI users in real-world listening situations. Further studies are needed to verify these findings in adult CI users and bimodal listeners.

Degree

Ph.D.

Advisors

Kirk, Purdue University.

Subject Area

Audiology|Neurosciences|Biomedical engineering

COinS