Parallel GMM Sub-Band Speaker Identification Over AMR-NB Codec

In this paper, we investigate an exploratory paradigm named parallel sub-band speaker recognition to improve speaker identification performance for mobile communication. The general principle of this concept is to split the whole frequency domain into several sub-bands on which statistical sub-recognizers based on Gaussian mixture models (GMM) are independently applied and then re-combined to yield a global score and a global recognition decision. The choice of the subband architecture and the recombination strategy are discussed. This technique had been shown to be robust for speech recognition when narrow band noise degradation occurs. We verify this robustness for the speaker identification task in the case of AMR-NB transcoded speech.