Contribution
The Importance of Ecological Validity in the Evaluation of Noise Reduction Systems with Hearing Aid Users
* Presenting author
Abstract:
In this study we evaluate the preference for different NR (Noise Reduction) systems with HA (Hearing Aid) users. We show how using ecologically valid speech + noise mixtures can lead to useful insights.Previous studies have shown that conversations in the wild mostly happen at positive SNRs (Signal to Noise Ratios). We selected a set of realistic noise signals, and we recorded speech + noise playback at different SNRs with HA prototypes. We analyzed the recordings with objective metrics, and we ran qualitative and quantitative tests with NH (Normal Hearing) listeners and HA users. We also compared synthetic mixtures to real speech in noise recordings.Objective metrics correlated well with preferences for NH listeners, but less so for HA users. Insights from qualitative research with users were instead key in defining relevant scenarios and refining algorithms. Ecologically valid acoustic scenarios and SNRs allowed HA users to relate to familiar listening situations and appreciate even subtle differences between NR systems. The comparison of synthetic mixes with real-world recordings revealed an important aspect which is typically overlooked in the evaluation of HA algorithms: real conversations in noise feature Lombard speech, which has an effect on both human perception and algorithms.