Speech Enhancement Using Maximum Likelihood and Maximum A Posteriori Detectors and Estimators

Clean speech detection in noisy signal is essential in estimation-based speech enhancement. In this paper, we derive optimum detector and estimator in a simultaneous manner. Whenever the proposed detector decides on the presence of speech, the derived optimum estimator extracts the target speech from noisy signal. We adopt Gaussian distribution for both clean speech and noise. Also, letting hit or miss function, we extract the formulas by maximum a posteriori (MAP) and maximum likelihood (ML) criteria. The main contribution of this work is the joint derivation of the optimum detector in the previous estimation-only methods. Simulation results show that the joint detector and estimator approach has a better performance than other estimation-based algorithms from both noise reduction and speech distortion viewpoints.