Evaluation of intelligibility of noisy whisper enhanced by compressive sensing

摘要

Noisy whisper contaminated by different types of noise are enhanced by compressive sensing method. The intelligibility performance of the enhanced whisper is measured using the short time objective index (STOI). To assess the effect of spectra density on compressive sensing, the real-valued discrete Gabor transform is used to transform the noisy whisper into the time-frequency domain. Experimental results show that, similar to conventional speech enhancement algorithm, the compressive sensing based speech enhancement method can not efficiently improve intelligibility of the whisper.