Using spectral constancy to encode temporal pitch and improve cochlear-implant melody perception.

Abstract

Despite good speech performance in quiet, melody perception remains a challenge to cochlear-implant (CI) users. Current CI pitch-encoding schemes significantly alter original signals, producing potentially detrimental effects on speech perception. The present study proposes to take advantage of spectral constancy to encode temporal pitch while minimizing signal distortion. Spectral constancy refers to unaltered timbre perception when the overall level of a signal is roved. For example, the vowel /a/ sounds the same independent of its overall level. The idea is to extract and encode pitch by increasing the gain of all channels proportionally and synchronously according to the pitch period. This scheme alters the temporal envelope of voiced sounds but preserves their spectral envelope in each time frame. It does not alter temporal envelope of unvoiced sounds, making it a good comprise between encoding pitch and preserving spectral and temporal envelopes. The new scheme has been implemented in a real-time speech processor and evaluated in nine nucleus CI users who identified ten melodies with minimal training and in a test protocol that randomized the stimulus presentation order. Compared with the standard continuous interleaved sampling strategy, the new scheme significantly improved CI melody perception by 10–20%.