Bayesian noise estimation in the modulation domain

Modulation domain has been reported to be a better alternative to time-frequency domain for speech enhancement, as speech intelligibility is closely linked with the modulation spectrum. Motivated by that, this paper investigates the use of modulation domain to model the noise density function. Results show that the modulation domain based Gamma density function better represents the noise density for all time-varying noise signals compared to the non-modulation domain. The modulation based Gamma density is then used to derive noise estimator via a Bayesian motivated MMSE approach. As the Gamma density closely matches the true noise spectrum in the modulation domain, the proposed noise estimator does not require bias compensation even for poor signal-to-noise ratio (SNR) conditions, i.e.,  ≤  5 dB. The proposed method yields better noise suppression compared to the state of the art methods and provides higher improvements.