Psychoacoustic model-driven spectral subtraction for monaural speech enhancement

In this paper, we investigate a psychoacoustic model-driven spectral subtraction framework for enhancement of noisy speech. In the proposed framework, the noisy speech spectrum is separated into six distinct and unevenly frequency-spaced subbands as per the psychoacoustic model of the human hearing system, and spectral over-subtraction is applied independently in each subband. The noise in each subband is estimated using an adaptive noise estimator that does not require a speech pause tracker. To compute and update the noise, the noisy speech power is adaptively smoothed using a smoothing factor controlled by a posterior SNR. The performance of the proposed framework is evaluated using SNR, segmental SNR (SegSNR), and PESQ scores for a variety of non-stationary and stationary noise environments at varying SNR levels. The experimental results show that the proposed framework outperforms various up-to-date speech enhancement technologies on three extensively used objective metrics assessments and speech spectrograms. © 2023, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.