In this paper, we consider speaker identificationfor the co-channel scenario in which speech mixture fromspeakers is recorded by one microphone only. The goal is toidentify both of the speakers from their mixed signal. Highrecognition accuracies have already been reported when anaccurately estimated signal-to-signal ratio (SSR) is available. Inthis paper, we approach the problem without estimating SSR.We show that a simple method based on fusion of adaptedGaussian mixture models and Kullback-Leibler divergencecalculated between models, achieves an accuracy of 97% and93% when the two target speakers enlisted as three and twomost probable speakers, respectively.
展开▼
