Spectral Magnitude Minimum Mean-Square Error Estimation Using Binary and Continuous Gain Functions

Abstract

Abstract-Recently, binary mask techniques have been proposed as a tool for retrieving a target speech signal from a noisy observation. A binary gain function is applied to time-frequency tiles of the noisy observation in order to suppress noise dominated and retain target dominated time-frequency regions. When implemented using discrete Fourier transform (DFT) techniques, the binary mask techniques can be seen as a special case of the broader class of DFT-based speech enhancement algorithms, for which the applied gain function is not constrained to be binary. In this context, we develop and compare binary mask techniques to state-of-the-art continuous gain techniques. We derive spectral magnitude minimum mean-square error binary gain estimators; the binary gain estimators turn out to be simple functions of the continuous gain estimators. We show that the optimal binary estimators are closely related to a range of existing, heuristically developed, binary gain estimators. The derived binary gain estimators perform better than existing binary gain estimators in simulation experiments with speech signals contaminated by several different noise sources as measured by speech quality and intelligibility measures. However, even the best binary mask method is significantly outperformed by state-of-the-art continuous gain estimators. The instrumental intelligibility results are confirmed in an intelligibility listening test.

Distributed delay and sum beamformer for speech enhancement in wireless sensor networks via randomized gossip

“…We further assume that the speech and noise are statistically independent. These signals are windowed and transformed into the frequency domain by applying the short-time discrete Fourier transform (DFT) leading to (1) where , and denote the noisy speech, target speech and noise DFT coefficient, respectively, at frequency-bin index , time-frame index and microphone . We assume the DFT coefficients to be independent in time and frequency, which allows us omit the time and frequency indices for brevity.…”

“…Speech enhancement algorithms can be categorized into two classes: single-channel and multi-channel techniques. Although single-channel algorithms can improve quality and have been shown to be able to improve speech intelligibility to some extent [1], improvements are generally modest as they can utilize only the spectral information [2]- [4]. Multi-channel speech enhancement algorithms have in theory the potential to improve the speech quality and intelligibility by using both spectral and spatial information about the speech and the noise sources [5], [6].…”

A CASA-Based System for Long-Term SNR Estimation

“…The IBM in the DFT domain is defined using (4), by comparing the energy (squared-magnitude) of clean speech and noise at each T-F unit. When estimating the binary mask, a recently proposed MMSE-based mask estimator is used [16]. We use Type-II binary masks as defined in [16] that minimize the spectral 'squared-magnitude' MSE.…”

“…When estimating the binary mask, a recently proposed MMSE-based mask estimator is used [16]. We use Type-II binary masks as defined in [16] that minimize the spectral 'squared-magnitude' MSE. It has the same form as the spectral magnitude MMSE mask derived in [16], except that the spectral squared-magnitude MMSE gain function is used in place of the gain function used in [16].…”

Online Monaural Speech Enhancement Based on Periodicity Analysis and A Priori SNR Estimation

“…Meanwhile, the algorithm derived and applied binary gain for the enhancement. The binary gain produces enhanced speech with low sound quality when compared to that produced by continuous (or soft) gain [10,11].…”