Bobby Gibbs, Ph.D.
Assistant Professor
UW Madison Dept of Communication Sciences and Disorders
Director, Auditory Encoding and Processing Optimization (AEPO) Lab
Converging Acoustic Evidence for Tailoring Speech Enhancement in Noise in Cochlear Implant Simulations
Speech in noise remains a top concern when listening through a cochlear implant (CI). Speech enhancement algorithms have shown limited success in noise. Improved noise mitigation requires a better understanding of what acoustic information is prioritized when listening through a CI in noise, and how acoustic utilization in noise is affected by the fidelity of initial neural encoding (the electrode-to-neural interface). The “bubble noise” paradigm provides an opportunity to test the hypothesis that acoustic utilization in noise depends on the electrode-to-neural interface. This talk will present analyses from bubble noise data when listening to consonants in noise. Bubbles are random regions of attenuation of an otherwise unintelligible masker that provide random glimpses of the phonemes. Test conditions involved vocoded stimuli with broad spread of excitation, vocoding with narrow spread of excitation, and unprocessed stimuli. I will present analyses from time-frequency importance functions (derived from correlating bubble regions with correct responses), error patterns, and acoustic intelligibility prediction metrics. The converging evidence from these analyses provides an initial roadmap for how CI speech enhancement in noise might be more tailored.