Modeling phoneme and open-set word recognition by cochlear implant users: a preliminary report.

Area of Science:

• Auditory Neuroscience
• Speech Processing
• Human-Computer Interaction

Background:

• Phoneme and word recognition are crucial for speech understanding, especially for individuals with hearing impairments using cochlear implants (CIs).
• Previous models often focus on acoustic-phonetic features, but the role of lexical information in open-set word recognition for CI users requires further investigation.

Purpose of the Study:

• To evaluate the predictive power of closed-set feature identification for phoneme and word recognition in an open-set task.
• To determine if incorporating a mental lexicon improves speech recognition predictions for cochlear implant users.

Main Methods:

• Utilized the Phoneme-to-Categorical Mapping (PCM) model to predict performance.
• Introduced the Spoken Phoneme and Lexicon Model (SPAMR) incorporating a mental lexicon to refine predictions.
• Collected and analyzed data from 7 adult cochlear implant users performing an open-set word recognition task.

Main Results:

• The PCM model accurately predicted phoneme identification but underpredicted word recognition.
• The SPAMR model, including a mental lexicon, significantly improved the match between predicted and observed word recognition performance.
• Findings support the hypothesis that lexical information is actively used by CI users during open-set spoken word recognition.

Conclusions:

• Closed-set feature identification is a useful predictor for phoneme identification in open-set tasks.
• Lexical knowledge plays a significant role in spoken word recognition for cochlear implant users.
• Future research will use individual CI user performance on psychophysical tasks to predict speech recognition more accurately.