Increasing spectral contrast enhancement worsens cochlear implant performance.

Evidence

In five CI users the new algorithm has been evaluated using a SMT task and a vowel identification task. The tasks were performed for SCE factors of 0 (no enhancement), 2 and 4. In general it seems that increasing the SCE factor produces a decrease in performance in both the SMT threshold and vowel identification. — Dijk, Pim van, Physiology, Psychoacoustics and…, 1821

Spectral resolution may be degraded by the broad electrical fields created in the cochlea when the electrodes are stimulated. In a recent study, the identification of spectrally smeared vowels and consonants was improved by spectral contrast enhancement (SCE) in a group of 166 normal hearing listeners (Alexander et al. 2011 ). Spectral contrast is defined as the level difference between peaks and valleys in the spectrum. In CIs, spectral contrast is degraded because of the limited number of stimulation electrodes and overlapping electric fields activating the nervous system through the bony structure of the cochlea. This might reduce the differences in amplitudes between peaks and valleys in the input making it more difficult to locate spectral dominance (i.e., formants) which provide crucial cues to speech intelligibility and instrument identification. — Dijk, Pim van, Physiology, Psychoacoustics and…, 1823

Spectral contrast is defined as the level difference between peaks and valleys in the spectrum. In CIs, spectral contrast is degraded because of the limited number of stimulation electrodes and overlapping electric fields activating the nervous system through the bony structure of the cochlea. This might reduce the differences in amplitudes between peaks and valleys in the input making it more difficult to locate spectral dominance (i.e., formants) which provide crucial cues to speech intelligibility and instrument identification. Loizou and Poroy 2001 showed that CI users need a higher spectral contrast than normal hearing listeners in vowel identification tasks. — Dijk, Pim van, Physiology, Psychoacoustics and…, 1827

A model of vowel identification has been used to investigate the effect of SCE on vowel identification. The model predicts that increasing the amount of SCE increases vowel identification accuracy. Based on these results we decided to use SCE factors 2 and 4 (equivalent to increasing the original spectral contrast of the spectrum by 3 and by 5 in a dB scale). The new SCE strategy has been evaluated in CI users. Results from a vowel identification task and a SMT task in five CI users show differences in vowel identification scores for different SCE factors. In general, it seems that SCE produces a detrimental effect in spectral modulation detection and vowel identification in CI users. These results are contrary to the model predictions. Previous studies in the literature give reasons to believe that spectral contrast enhancement would result in a benefit for the chosen tasks. It is possible that spectral valleys are attenuated too much and relevant information required by the CI users to understand speech is lost. These effects are not taken into account by the MPI model, and this could explain the contradictory results between experiments and CI users and modelling results. — Dijk, Pim van, Physiology, Psychoacoustics and…, 1886

Categorization

Category: Contrarian or Surprising Claims

• XREF: Connects to the broader literature on spectral resolution and CI performance; challenges the intuitive premise that sharper spectral peaks help speech perception.
• DEFINE: Defines spectral contrast as peak-to-valley level difference in the spectrum.
• QUESTION: The possibility that spectral valleys are over-attenuated and lose speech-relevant information is a testable hypothesis about CI signal processing worth chasing.
• SHARE: Worth sending to anyone working in auditory perception or cochlear implant signal processing.

Related

• Spectral valley over-attenuation may explain why contrast enhancement harms CI users (similarity 0.89)
• Increasing spectral contrast enhancement worsens vowel identification in CI users (similarity 0.89)
• Cochlear implant spectral smearing flattens formant peaks needed for speech recognition (similarity 0.87)
• Spectral contrast degradation in CIs impairs formant detection for speech (similarity 0.85)
• Over-attenuating spectral valleys may remove speech information CI users need (similarity 0.82)
• Spectral resolution closely predicts vowel and consonant recognition in CI users (similarity 0.82)
• Cochlear implants give good speech recognition in quiet but poor performance in noise (similarity 0.80)
• More hearing aid compression channels measurably degrade vowel perception in hearing-impaired listeners. (similarity 0.80)
• More compression channels reduce spectral contrast and vowel recognition (similarity 0.79)
• Deactivating cochlear implant electrodes harms spectral ripple resolution but spares speech perception (similarity 0.79)
• Spectral modulation thresholds quantify the smallest detectable spectral contrast (similarity 0.79)
• Spectral ripple thresholds weakly predict vowel identification in cochlear implant users (similarity 0.79)
• CIE electric-field spread flattens spectral contrast, obscuring formants needed for speech (similarity 0.78)

Broader Connections

• Increasing spectral contrast does not improve spectral modulation discrimination thresholds #corroborates (0.85) — Dijk, Pim van, Physiology, Psychoacoustics and…, 1872
• More compression channels reduce spectral contrast and vowel recognition #corroborates (0.72) — Jennifer Lentz, Psychoacoustics- Perception of Normal…, 1730
• Concentrated spectral energy damages the ear more than broadband noise of equal level #exemplifies (0.70) — David Howard, Jamie Angus, coustics and Psychoacoustics, 676