New approach to speech intelligibility testing

New approach to speech intelligibility testing
30 august 2017


Measuring the effect of e.g. noise reduction is traditionally done using speech-intelligibility-in-noise tests. These measure percent correct at a given signal-to-noise ratio (SNR).

Usually such tests are done at a SNR below 0 dB (signal ≤ noise) because this is where they are most sensitive. However, a recent study from 2012* of SNR levels in ecological listening situations (e.g. kitchen, babble and car) shows that the SNRs in the ten most common listening situations of hearing aid users are all well above 0 dB.

This indicates that in order to test the eventual benefits of signal processing interventions in an ecologically valid listening situation, testing should be done within the SNR range of +5-15 dB. This is also the SNR range where measuring eventual benefits goes beyond the capability of traditional speech-in-noise tests. Consequently, new measures of testing are needed, and a useful measure is found to be: Cognitive Spare Capacity.

Cognitive Spare Capacity refers to the residual capacity after successful speech perception. In a quiet situation where speech input comes through completely undisturbed, this is understood effortlessly with help from our highly over-learned automatic speech recognition system. In a more difficult listening situation we need to concentrate more in order to understand what is being said, and we automatically draw more on our working memory. This puts more strain on our brain, leaving less spare capacity, e.g. for memory. We argue that, compared to normal hearing persons, hearing impaired persons rely much more on effortful working memory resources to understand what has been said, because of the extra distortions imposed by their hearing impairment.

Therefore tests which probe the degree of load on working memory resources are useful for evaluating the benefit of hearing aids.

In a recent study from 2013** residual capacity was defined as the number of words recalled after successful listening to a number of HINT sentences. This was followed up in a new study*** with 25 hearing impaired test subjects. All participants demonstrated close to 100% correct speech intelligibility when exposed to a four talker babble noise around + 7-9 dB SNR. At the same SNR level, a hearing aid noise reduction scheme improved memory recall by about 10%.

Thus, this kind of memory recall test is a possible candidate for assessment of hearing aid functionality in ecologically relevant (positive) SNRs.


* Smeds et al. 2012

** Ng et al. 2013

*** Thomas Lunner1,2, Elaine Ng2, Maria Grube Sorgenfrei1,3, Marianna Vatti1, Renskje Hietkamp1, Dragan Gusatovic3, Graham Naylor1

1. Eriksholm Research Centre, Oticon A/S, Snekkersten, Denmark

2. Dept. of Behavioural Sciences & Learning, Linköping University, Sweden

3. Oticon A/S, Smørum, Denmark