BrainHearing: Attending to competing voices

BrainHearing: Attending to competing voices
5 august 2017


At social situations, we are used to attending to more than one talker at the same time. This task can be performed by humans in very adverse acoustic conditions, in which technology such as automatic speech recognition systems will fail completely.

In order to perform this task, we must rely on acute peripheral hearing, binaural combination of the neural signals from both ears and finally the brain to decode one or multiple messages (Recognize and Separate), depending on the attention (Focus) of the listener. This process goes on real-time, using the available means – language knowledge, context, previous knowledge etc – and requiring considerable cognitive effort. We will refer to this problem as the ‘competing voices problem’.

The effects of hearing loss on speech recognition are well-known: discrimination loss and a raised speech reception threshold in noise. On top of this are the gradually reduced cognitive skills associated with hearing loss. So the competing voices problem would seem impossible to handle for an older hearing-impaired person.

At Eriksholm Research Centre, a competing voices test is being developed to assess performance in the scenario where two equally important talkers are attended to. Such at test is not yet available in the research community. The first version of the test from 2014* has recently been evaluated on seven elderly hearing-impaired listeners and presented at the International Hearing Aid Conference (IHCON 2014) at Lake Tahoe, California.

When listening to sentences on a continuous story (read by a male and a female), the speech recognition of the sentences were close to 100%, and questions to the continuous story could also be answered with some errors. While listening to dual simultaneous sentences (read by a male and a female), the scores were in the 10 – 60% word score range for both sentences, depending on the exact task: Listeners were asked to repeat both sentences or only one the them cued from the screen either before or after the playback of the sentences (see figure). They would thus rely on their Recognize, Separate and Focus skills. The listeners judged the task to be difficult, but not impossible and could also relate to similar challenges in their daily life.

These preliminary results demonstrate that even with age and impaired hearing the brain is still able to complete the task of attending to competing voices. Detailed knowledge of the important spatial, spectral and temporal cues and how to preserve them in a hearing aid is vital for the best possible listening experience in competing voice situations.

Find the IHCON poster (and other Eriksholm contributions) here:


*Bramsløw, L., Vatti, M., Hietkamp, R., & Pontoppidan, N. H. (2014). Design of a competing voices test. In International Hearing aid Conference (IHCON) 2014. Lake Tahoe, CA, USA.