Is explicit formant encoding useful for speech perception with cochlear implants

Earlier generations of cochlear implant (CI) sound processing strategies incorporated explicit encoding of formant frequencies. Many of these CI strategies have yielded higher intelligibility outcomes for CI users due to the pure spectral-based approaches used in contemporary processor today, such as: CIS, ACE, Hi-Res, etc. The aim of the study has been to assess intelligibility due to possible loss in formant frequency encoding and its influence on channel selection in diverse acoustic conditions. Formant accuracy from four computer-aided estimation techniques are compared against hand-marked frequencies derived from phoneme-level transcription with sentences from the IEEE and AzBio databases in three noise types at different SNRs: (a) babble, (b) speech-shaped, and (c) noise and reverb combination. CI speech intelligibility outcomes of the proposed strategy using explicit formant frequency encoding was compared to ACE, the standard energy-based ‘n-of-m’ approach. A relationship between the estimation techniques and differences in channel selection was systematically assessed. Results suggest that minor loss of formant precision does not yield statistically significant differences in resulting speech intelligibility. However, availability of accurate formant cues may help with intelligibility gains in challenging noise such as babble.