Abstract: A method for discriminating noise from signal in a noise-contaminated signal involves decomposing a frame of samples of the signal into decorrelated components, and using a difference between probability distributions of the noise contributions and the signal contributions to identify signal and noise. A Gaussian distribution is used to determine whether the components are only noise whereas a Laplacian distribution is used to determine whether the components contain the signal. Such discrimination may be used in speech enhancement or voice activity detection apparatus.

Abstract: System and methods are provided based on optimization of the weighted log-likelihood. These systems are able to efficiently track dominant sinusoidal components of a real signal in Gaussian noise, provided that the number of the components is known. The algorithm is implemented using simple parallel building blocks involving narrow-band filters that are adaptively self-tuned around the frequencies of the signal components. The algorithm has low computational complexity and provides high estimation accuracy and is also able to track chirp signals. The algorithm is flexible enough to be adjusted to operate in different environments such as for speech signals, by selecting a proper window function. Simulation results confirm that the proposed algorithm is reliable in tracking the frequencies as well as in estimation of the amplitudes of the components. In a chirp environment, the algorithm is able to recognize some frequency cross-overs as long as the amplitudes are different enough around the cross-over moment.