ISCA Workshop on Plasticity in Speech Perception (PSP2005)
Senate House, London, UK
Speech recognition with a cochlear implant is impaired by tonotopic misalignment between the frequency band transmitted by an electrode and the characteristic frequency of the location of that electrode (Skinner et al., 2002: JARO 3 332-350). Experiments with normally-hearing subjects who listen to simulations of the information provided by cochlear implants have also shown that speech recognition is poor when signals are spectrally shifted to simulate tonotopic misalignment (Rosen et al., 1999: JASA 106, 3629-3636). Although subjects can partially adapt to such signals (e.g. Rosen et al., 1999), previous studies have not distinguished between improvements that result from training and improvements that result from repeated testing.
The present experiment investigated the effectiveness of an unsupervised high-variability lexical training regime in 32 normally-hearing subjects who listened through an 8-channel vocoder simulation of the information provided by a cochlear implant. A 3-mm basalward tonotopic misalignment was simulated. The training task was a 2-alternative forced-choice procedure. On each trial, subjects chose which of two words displayed orthographically on a computer screen matched a target word. Feedback on accuracy was given. In the auditory version of the training task, targets were presented acoustically. In the control version, targets were presented orthographically masked by visual noise. Following baseline testing, all subjects received a session of 1200 trials of auditory training and a session of 1200 trials of control training in a counter-balanced order. Half the subjects received High-Variability auditory training, in which targets were recorded by 10 talkers (4 males, 4 females, 2 female children). The other half received Single- Talker auditory training, in which targets were recorded by a single male. Effects of training were assessed by measuring the percentage of key words correctly identified in IEEE sentences recorded by 10 talkers.
Auditory training was more effective than control training: control training produced an improvement of 4.1% (95% c.i. 2.0-6.2%), while auditory training produced an improvement of 9.4% (95% c.i. 7.4- 11.4%). There was a trend for High-Variability auditory training to be associated with greater improvements (mean improvement 11.5%, 95% c.i. 8.7-14.2%) than Single-Talker auditory training (mean improvement 7.4%, 95% c.i. 4.6-10.1%).
These results show that auditory training is effective, even when exposure to the training vocabulary and the test materials is controlled. Although the benefits of (a single session of) training were small, the results are encouraging as they indicate that an unsupervised computer-based lexical training regime can improve the ability to recognise spectrally distorted speech.
Bibliographic reference. Stacey, Paula / Summerfield, Quentin (2005): "Auditory-perceptual training using a simulation of a cochlear-implant system: a controlled study", In PSP2005, 143-146.