Contribution
Predicting Speech Recognition of Hearing-Impaired Listeners in Virtual Acoustic Scenes and Standard Conditions: A Validation of FADE
* Presenting author
Abstract:
Speech recognition in hearing-impaired individuals is typically evaluated in simplified, anechoic conditions, but real-life scenarios involve acoustic environments with varying numbers and positions of sound sources and reverberation. A new database has recently been established to address this gap, including speech recognition measurements of elderly normal-hearing and (aided) hearing-impaired listeners in standard conditions (S0N0 and S0N90) and conditions reflecting real-life scenarios using virtual acoustic scenes (living room, pub, underground station). To better understand the performance mismatch between standard and simulated complex conditions, speech recognition thresholds (SRTs) of individual listeners were predicted using the Simulation Framework for Auditory Discrimination Experiments (FADE) with the binaural stage CAIN. This auditory model was validated for predicting individual SRTs in virtual acoustic scenes, with R² correlation values between measured and predicted SRTs of 0.67 (symmetric living room), 0.74 (pub), 0.76 (underground station), and 0.81 (asymmetric living room). However, FADE tends to overestimate speech recognition performance in more reverberant scenes, highlighting its limitations. FADE can help evaluate individual speech recognition abilities in different environments, with potential applications in selecting individualized hearing aid settings. Its limitations should be considered in conjunction with its strengths in predicting the effect of various maskers and hearing impairment on speech recognition.