Speaker
Description
Background. Advances in acoustic feature extraction and machine learning enable conversation detection in real-world settings. This provides unique insights into the acoustic characteristics associated with communication. This study assessed the influence of hearing acuity on conversation frequency, own speech levels, and speaking time.Methods. Acoustic features and Ecological Momentary Assessment (EMA) data were used to detect daily conversations, without accessing raw audio, in 90 participants (aged 40-73 years, mean 59 years) with varying levels of hearing acuity (Pure Tone Average thresholds [PTA] of the better ear, which ranged between -5 and 94 dB HL, mean 28 dB HL). Using the binaural olMEGA microphone system and a random forest classifier trained on EMA-prompted self-reported conversation periods, the likelihood of conversation was predicted throughout the day. The participants’ voice was separated from ambient sound and own speech level was assessed. The relationships between PTA and hearing loss (yes/no), and conversation characteristics (frequency, speaking time, speech level) were then assessed. Results. Using leave-one-subject-out cross-validation, the conversation classifier achieved a mean weighted F1 score of 0.73. Conversation likelihood was associated with noisier environments and hearing acuity. During conversations, speech levels increased with ambient noise. Speaking time was also associated with ambient sound levels, and this effect depended on PTA. Conclusions. This study shows the potential of combining acoustic and EMA data to assess daily communication. The inability to separate background sound from conversation partners’ speech and the short monitoring periods remain a key limitation. Future research should address these and explore individual communication strategies to better understand how hearing acuity affects social interactions.