Guided stepwise training

Guided stepwise training enables hearing-impaired listeners to perform perceptual audio evaluations on stimuli containing artefacts. Without this training the subjects failed on almost all tested conditions in a pilot study.

What is perceptual audio evaluation?

It can be assumed that future hearing-aid algorithms or assistive listening systems may include signal compression. This raises the question as to whether artefacts will be audible to the hearing-aid user. One method for answering this question is perceptual evaluation, which is common in audio and telecommunication fields¹, using normal hearing subjects. These subjects are trained prior to testing, typically with the aid of a listening panel, where a glossary of sound quality is agreed upon.

Why guided stepwise training?

We performed a pilot study in the form of a listening experiment with music and speech stimuli, processed to generate different degrees of non-linear artefacts. Six subjects with mild-to-moderate, sloping hearing losses were tested after some initial - and ample - training. Results were contradictory, but seemed to indicate that they were unable to detect differences in sound quality.

Given the ample training, we speculated that the subjects might have misunderstood the ABX task. Furthermore, the lack of familiarisation might explain some of the results. Apparently, there was a need for a targeted training program, keeping in mind that the impaired hearing system differs markedly from the normal hearing system and shows large, individual and highly unpredictable variation. Therefore, an individual training procedure was called for rather than a listening panel.

What is guided stepwise training?

A stepwise training procedure was developed for perceptual audio evaluation targeting hearing-impaired listeners. This was inspired by Bech and Zacharov¹. A key issue in the training procedure is priming, i.e. directing people’s attention to the artefacts under study. The basis for the procedure is errorless learning, where only one step is taken at a time with an increasing level of difficulty.

The method used is the ABX paradigm², where the subject is presented with two known samples, A and B (one being the reference, the other processed, randomly assigned) together with a third (X) that is the same as either A or B. The subject needs to identify which sample (A or B) corresponds to X.

Does this procedure work?

Using the developed stepwise training procedure in an experiment with the same type of stimuli as before and with the same test subjects, the results showed that differences in sound quality now were detected in the 2-5 kHz region. This was true not only for the Speech 1 stimuli (which the subjects had been subjected to before in the pilot study and thus had received more training in), but also for the new Music 3 and Music 4 stimuli. These results lead us to conclude that using hearing-impaired listeners for perceptual audio evaluation demands carefully designed training

Further reading

¹ Bech, S. and Zacharov, N. (2006). Perceptual Audio Evaluation - Theory, Method and Application. West Sussex: Wiley.

² Clark, D. (1982). High-Resolution Subjective Testing Using a Double-Blind Comparator. Journal of the Audio Engineering Society, 30(5), p. 330-338.

Hietkamp RK, Andersen MR, Lunner T (2009) Perceptual audio evaluation by hearing-impaired listeners – some considerations on task training. In: Buchholz J, Dalsgaard JC, Dau T, Poulsen T (Eds.) Binaural Processing and Spatial Hearing. Proceedings of the Second International Symposium on Audiological and Auditory Research, ISAAR, p. 487-496.