Addressing Channel Mismatch through Speaker Discriminative Transforms

This paper presents a discriminative criterion applied to Gaussian mixture models (GMMs) to reduce handset mismatch. The criterion is related to the log-likelihood-ratio (LLR) scoring approach commonly used in GMMs for speaker recognition. The algorithm attempts to perform a direct mapping of features from one channel type to an assumed undistorted target channel but with the goal of maximizing speaker discrimination using the transform. The transform attempts to maximize the posterior probability of a group of speaker models given their corresponding speech observations recorded on a different channel