Selecting the right speech material for speech-in-noise tests

Reading Time: 3 Minutes
30.01.2019

Our research results presented at the Speech-in-Noise Workshop in Ghent, Belgium, January 2019

The Speech-in-Noise (SpiN) Workshop is a unique occasion to update and expand my knowledge about different aspects of speech-in-noise. The conference covers a wide range of topics that are related to how we understand speech in adverse conditions and the potential applications of these research findings. It is also a chance to share our own results and get feedback from researchers working in different fields.

Our research

At this workshop, we presented a poster with some results of our last clinical trial. Our research investigated the effect of speech material on the measured benefit of amplification. It is crucial for any hearing aid or implant manufacturer to test, select, and verify the performance of different solutions, including different settings in an algorithm, different fitting strategies, or different devices. The results of our clinical tests have a major impact on the quality of the released product because they will determine which prototype or solution will be implemented in a commercial product. Therefore, it is also important to have the right measurement tools to get reliable results.

Measurement procedures

For many hearing aid features, speech-in-noise tests are an accepted evaluation method. There are two main types of measurement procedures: the fixed method where the signal-to-noise ratio (SNR) is kept constant during the test and the adaptive procedure where the SNR is updated to target a pre-defined performance level, which is usually 50%. Adaptive procedures are popular as the speech reception threshold (SRT), which is the SNR where 50% is correctly repeated, can be found in less than 5 minutes. However, the choice of the speech material must be clearly defined as it might impact the measured results.

Speech material

Speech material can be classified into two main categories: a) matrix sentences with a fixed grammatical structure which provides a uniform test material once the training phase is done, and b) everyday sentences that introduce more variability in terms of structure and word predictability. Current research as well as our own results have shown that everyday sentences must be presented at a more favorable SNR to achieve the same performance. So, one possible consequence is that when we measure the aided SRT at different SNRs then we might find different amplification benefit when the device has an SNR dependent signal processing; in other words, noise reduction and directionality give more benefit in more challenging listening environments or negative SNRs.

Observations

Our research evaluates the amplification benefit as a function of the speech material (matrix vs. everyday sentences), the unaided speech-in-noise understanding score, and different audiometric factors. A first interesting observation is that the benefit of amplification strongly depends on the unaided SRT. More amplification benefit was shown for listeners with high unaided SRTs (difficulties for speech-in-noise). It means that hearing aid users with good (or low) SRTs will show relatively less benefit in the aided condition. The second major finding is that the amplification benefit is more important when it is measured with a matrix test. While this test has less variation in structure and words, it might be also less representative of natural conversation.

Conclusions

Our conclusions are that aided speech-in-noise scores cannot be analyzed without knowing the individual unaided performances and that these scores should always be interpreted regarding all the parameters of the test setup, like the speech material. Comparison between scores obtained with different test setups must be carefully analyzed before any attempt at generalization. Sharing our results is an interesting and necessary process in order to improve the quality of our test protocol and data analysis. It is a good opportunity to better understand psychological or linguistic aspects related to speech understanding and associated with our work.

If you would like to read our poster presentation, you can find it in the Download Center using this link.

If you would like to read another SpiN Workshop blog post, click here.

Bernafon at the SPiN Workshop in Gent, January 2019
Bernafon at the SpiN Workshop in Ghent, Belgium, January 2019

 

About the author:

Christophe Lesimple
Christophe Lesimple
Christophe is a Clinical Research Audiologist and has worked for Bernafon since 2011. He contributes to various aspects of development like working on concepts, running clinical trials, and analyzing data. Besides his activities with Bernafon, he teaches research methods and statistics at the University of Lyon. In his private time, Christophe likes to play music and volunteer for a hearing impaired association.

 

Comments