Improving speech intelligibility in perceptual wavelet packet-based speech coding for cochlear implants

Abstract

Speech intelligibility for noise reduction algorithms which were integrated into perceptual wavelet packet-based speech coding strategy in cochlear implant (CI) processors was investigated in this study. The noise reduction algorithms including time-adaptive wavelet thresholding (TAWT) and time-frequency spectral subtraction (TFSS) were selected for this study due to simple and suitable for real-time implementation. The experiments were compared without and with noise reduction algorithms for fourteen normal-hearing (NH) listeners. The speech sentences were corrupted by babble noise in different levels of signal-to-noise ratio (SNR) (i.e. 0, 5 and 10 dB). The experimental results showed that vocoded noisy speech with TAWT and TFSS provided higher intelligibility at 0 and 5 dB SNR but slightly lower intelligibility at 10 dB SNR when compared to vocoded noisy speech. CI listeners may benefit more than NH listeners in further study.