Artificial stereo extension based on hidden Markov model for the incorporation of non-stationary energy trajectory

In this paper, an artificial stereo extension method is proposed to provide stereophonic sound from mono sound. While frame-independent artificial stereo extension methods, such as Gaussian mixture model (GMM)-based extension, do not consider the correlation of energies of previous frames, the proposed stereo extension method employs a minimum mean-squared error estimator based on a hidden Markov model (HMM) for the incorporation of non-stationary energy trajectory. The performance of the proposed stereo extension method is evaluated by a multiple stimuli with a hidden reference and anchor (MUSHRA) test. It is shown from the statistical analysis of the MUSHRA test results that the stereo signals extended by the proposed stereo extension method have significantly better quality than those of a GMM-based stereo extension method.