Real and imaginary modulation spectral subtraction for speech enhancement

In this paper, we propose a novel spectral subtraction method for noisy speech enhancement. Instead of taking the conventional approach of carrying out subtraction on the magnitude spectrum in the acoustic frequency domain, we propose to perform subtraction on the real and imaginary spectra separately in the modulation frequency domain, where the method is referred to as MRISS. By doing so, we are able to enhance magnitude as well as phase through spectral subtraction. We conducted objective and subjective evaluation experiments to compare the performance of the proposed MRISS method with three existing methods, including modulation frequency domain magnitude spectral subtraction (MSS), nonlinear spectral subtraction (NSS), and minimum mean square error estimation (MMSE). The objective evaluation used the criteria of segmental signal-to-noise ratio (Segmental SNR), PESQ, and average Itakura–Saito spectral distance (ISD). The subjective evaluation used a mean preference score with 14 participants. Both objective and subjective evaluation results have demonstrated that the proposed method outperformed the three existing speech enhancement methods. A further analysis has shown that the winning performance of the proposed MRISS method comes from improvements in the recovery of both acoustic magnitude and phase spectrum.

► We perform real and imaginary spectral subtractions in modulation frequency domain.
► Our method enhances both magnitude and phase speech spectra from noise.
► Our method showed superior outcomes in segmental SNR, PESQ, and averaged ISD.
► Our method showed superior outcome in mean preference score of listening evaluation.
► We analyze the factors contributing to our method’s winning performance.