Hearing-integrated bilingual speech corpus: A French-English corpus including hearables for studying speech production under challenging listening conditions

Communication challenges are exacerbated by noise, particularly for individuals with hearing impairment who may also have ear occlusion from hearing protection devices in occupational settings. These combined effects on speech production are understudied, despite auditory feedback being crucial for speech motor control and thus effective communication. This paper introduces Hearing-Integrated Bilingual Speech Corpus, a comprehensive database for examining speech production across varying levels of noise, ear occlusion, and hearing thresholds. We recruited 49 participants [19 with at least one frequency at ≥ 20 dB hearing level (dBHL)] who completed sentence reading, sustained vowel production, and picture description tasks. We demonstrate the database's utility by analyzing speech level responses in the sentence-reading task. We investigated all three factors categorically and additionally modeled hearing impairment as a continuous variable using pure-tone average (PTA). Key results showed that ear occlusion led to increased speech level, but the relationship exhibited non-linearity. Additionally, preliminary findings revealed that participants with PTA > 15 dBHL spoke louder overall, and a categorical shift occurs around PTA = 15 dBHL, where individuals with greater hearing impairment became less reactive to noise under high occlusion conditions. The considerable individual variability challenges categorical groupings of all investigated variables and highlights the need for individualized modeling approaches.