Speech recognition under masking: Age, hearing, and machine learning classification

In this study, we seek to empirically evaluate whether maskers can be categorically grouped into energetic and informational using machine learning classification techniques. The study further aimed to examine how age and hearing ability affect speech reception thresholds (SRTs) using different speech materials and masker types (energetic vs. informational). PARTICIPANTS INCLUDED YOUNG AND OLDER INDIVIDUALS WITH NORMAL HEARING, AS WELL AS OLDER INDIVIDUALS WITH HEARING IMPAIRMENTS (HI). TWO SPEECH MATERIALS TARGETING DIFFERENT SRTS WERE USED. MACHINE LEARNING CONFIRMED THE CLASSIFICATION OF NOISES INTO ENERGETIC AND INFORMATIONAL MASKERS: Linear mixed-effects modeling predicted signal-to-noise ratios (SNRs) in the Swedish Hearing in Noise Test (HINT) and Hagerman test based on group, SRTs, and Masker Type. Results showed that younger and older adults performed similarly in the HINT with informational maskers, which had significantly lower SNRs. Older HI individuals had higher SNRs in both Masker Types. Differences were noted in speech recognition criteria across the three groups in both tests. The findings indicated that machine learning supported the theoretical categorization of maskers. Furthermore, older and younger normally-hearing adults performed similarly regardless of speech recognition criteria or Masker Type, while HI individuals performed worse in all conditions. This suggests that normal-hearing older adults may use strategies to match younger adults' performance, and that peripheral hearing, rather than age, significantly impacts speech recognition in various conditions.