Evaluation of Thai speech intelligibility based on noise reduction techniques for cochlear implants

Abstract

Speech intelligibility improvement in noisy environments for cochlear implant (CI) users is a difficult task which has encountered numerous challenges, such as differences between languages. Most studies have been carried out using the English language. The objective of the study was to evaluate Thai speech intelligibility based on noise reduction (NR) algorithms as a pre-processing approach in CIs. Two NR algorithms, namely multi-band spectral subtraction (MBSS) and Weiner filter (WF) algorithms, were chosen. Monosyllabic and bisyllabic Thai words were corrupted by different noises (speech-shaped noise and babble noise) at SNRs of 0, 5, and 10 dB. Then, the noisy speech was enhanced using NR algorithms. The enhanced speech was fed into a n-of-m coding strategy to synthesize vocoded speech. The vocoded speech was evaluated by ten normal-hearing (NH) listeners. The experimental results showed that intelligibility performance was slightly improved for the MBSS algorithm and considerably improved for the WF algorithm in some conditions. The intelligibility performance of bisyllabic words was significantly more improved than that of monosyllabic words in all conditions.