We examined the convergence of simultaneous fNIRS, EEG, and behavioral outcome measures in evaluating the impact of deep neural network noise reduction (DNN-NR) versus traditional noise reduction (SPIN) methods in simulated cocktail party conditions. Twenty-nine experienced hearing aid users completed an 8-week crossover field trial using Phonak Audeo Sphere Infinio devices. In a speech stream segregation task (SSST), cognitive load was modulated by varying lexical frequency of target words presented in a diffuse complex (café, music, steady-state noise) background. Simultaneous outcome measures included behavioral performance accuracy and response latency, 31-channel functional near-infrared spectroscopy (fNIRS), and 32-channel electroencephalography (EEG). A live evaluation of auditory preference (LEAP) assessed real-world listening preference during group conversation. DNN-NR significantly improved word identification and reduced response latency compared to SPIN. Lexical difficulty affected performance only in SPIN, suggesting that DNN-NR mitigates cognitive load. fNIRS showed reduced oxygenation with DNN-NR in prefrontal and temporal cortical regions, indicating reduced neural demand for cognitive, language and auditory processing. EEG revealed increased alpha and gamma activity linked to enhanced speech tracking and selective attention. In the LEAP, participants preferred DNN-NR by a 3.5:1 ratio.

Learning Objectives:

1. Compare and contrast traditional and deep-neural network (DNN)-based noise reduction strategies in hearing aids.
2. Interpret multimodal neural data (fNIRS and EEG) to understand how different noise reduction algorithms influence cognitive and auditory processing during speech perception tasks.
3. Describe how to use and interpret listening preference data from live conversational settings to assess the ecological validity and real-world benefits of DNN-based noise reduction in hearing aid technology.