An Intelligent Directionality System for Hearing Aids Incorporating Deep Neural Networks: Improving Speech Understanding While Preserving Spatial Awareness.

Background

Directionality algorithms are fundamental to modern hearing aids, enhancing speech perception in noisy environments by improving the signal-to-noise ratio. Conventional adaptive approaches rely on heuristic optimization criteria and often improve speech understanding from specific directions, typically the front, at the expense of spatial awareness. A deep neural network (DNN)-based directionality system was developed that dynamically optimizes spatial filtering through a data-driven framework capable of representing complex acoustic scenes. The system intelligently enhances target speech while preserving awareness of environmental sounds.

Methods

The proposed DNN-based directionality system was evaluated against conventional directionality algorithms by assessing word recognition performance, perceived speech clarity, and listener preference across diverse acoustic scenarios. To complement subjective measures, an artificial intelligence (AI)-based objective intelligibility metric was additionally developed using automated speech-to-text analysis, enabling scalable and consistent benchmarking across devices.

Results

Results across the behavioral and objective assessments indicate improved speech intelligibility outcomes together with maintained access to environmental sounds. Mean SRT50 improved by 1.4 dB relative to legacy directionality for target speech presented at 90° and by 4.7 dB relative to an omnidirectional pattern when frontal target speech was accompanied by a rear interfering talker. Spatial adaptation further improved environmental sound audibility in three out of four conditions and significantly improved the detection threshold for speech presented at 135° by 2.42 dB.

Conclusions

These results show that the system can preserve speech originating away from the front while using acoustic context and spatial filtering to attenuate competing speech that interferes with a frontal conversation.