Relative contributions of temporal and spectral cues for Mandarin and Cantonese tone recognition

Abstract

Despite good speech perception performance in quiet, pitch perception remains a challenge for cochlear-implant users. The current vocoder-based processing technique in cochlear implants preserves temporal envelope and coarse spectral information. This information, however, is insufficient to support high-level performance for tone perception in tonal languages (e.g., Mandarin and Cantonese), particularly in noise. Mandarin has four lexical tones and Cantonese has six lexical tones. Each lexical tone has different fundamental frequency contours. The temporal envelope of the signal also differs among different tones. In this talk, we will first discuss the relative contributions of temporal and spectral envelope and fine structure cues for Mandarin and Cantonese tone perception. We will then describe a new signal processing algorithm developed at the Chinese University of Hong Kong that aims to enhance temporal periodicity cues for Cantonese tone perception in noise. We will present findings from normal-hearing and cochlear-implant listeners, which demonstrate a significant improvement of Cantonese tone perception in noise with this new algorithm.