Spatial AI improves speech intelligibility for hearing aid wearers in challenging multitalker noise.

OBJECTIVE: We evaluated the Spatial AI model on the Fortell AI hearing aids to see if it improves speech intelligibility in single-azimuth multitalker babble relative to a commercially available control hearing aid employing neural-network-based noise-reduction. DESIGN: In a pre-registered, double-blind, randomised controlled trial, hearing aid users completed a QuickSIN test wearing the Fortell and control devices. Both devices were set to maximise directionality and noise reduction. Objective metrics were also calculated. SAMPLE: Thirty adults with mild to moderately-severe bilateral sensorineural hearing loss. RESULTS: The investigational hearing aid showed significant improvements in speech intelligibility, with odds ratios of 6.0, 10.6 and 18.9 across SNRs of +6, 0, and -6 dB (all p ≪ 0.0001). The SNR-50 improved by 9.2 dB and all participants identified more words correctly using the investigational device. Lab measurements showed an SNR improvement over unaided of 2.3 dB with the control device and 15.5 dB with the investigational hearing aid. HASPI v2 showed a similar level of SNR improvement. CONCLUSIONS: Spatial AI yields substantial intelligibility benefits in a controlled single-azimuth multitalker configuration. The effect sizes greatly exceeded established thresholds for clinical significance, suggesting that the algorithm may help address the cocktail party problem for hearing-impaired listeners.