Speech wideband extension based on Gaussian mixture model

To decrease the spectral distortion of highband envelope, the function of spectral distortion and mutual information which is between feature vector and highband envelope is studied, and an extended Gaussian Mixture Model (GMM) wideband extension algorithm is proposed based on the research. The feature parameters which have higher mutual information with highband envelope are selected to constitute feature vector, and the GMM is adopted to compute the joint probability density of the feature vector and highband envelope. Then the highband envelope is estimated via the posterior probabilities computed from the model parameters estimated by Expectation-Maximization (EM) Mgorithm. The experimental results show that the spectral distortion is inferior to the Mgorithm, such as the traditional algorithm based on GMM, by 0.3 dB and the number of frames with spectral distortion over 10 dB sharply reduced over 50%.