Text Independent Speaker Identification using Integrating Independent Component Analysis with Generalized Gaussian Mixture Model

Recently much work has been reported in literature regarding Text Independent speaker identification models. Sailaja et al (2010)[34] has developed a Text Independent speaker identification model assuming that the speech spectra of each individual speaker can be modeled by Mel frequency cepstral coefficient and Generalized Gaussian mixture model. The limitation of this model is the feature vectors (Mel frequency cepstral coefficients) are high in dimension and assumed to be independent. But feature represented by MFCC’s are dependent and chopping some of the MFCC’s will bring falsification in the model. Hence, in this paper a new and novel Text Independent speaker identification model is developed by integrating MFCC’s with Independent component analysis(ICA) for obtaining independency and to achieve low dimensionality in feature vector extraction. Assuming that the new feature vectors follows a Generalized Gaussian Mixture Model (GGMM), the model parameters are estimated by using EM algorithm. A Bayesian classifier is used to identify each speaker. The experimental result with 50 speaker’s data base reveals that the proposed procedure outperforms the existing methods.

@article{Ramaligeswararao2011,
  title     = {Text Independent Speaker Identification using Integrating Independent Component Analysis with Generalized Gaussian Mixture Model},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {2},
  number    = {12},
  year      = {2011},
  publisher = {The Science and Information Organization},
  author    = {N M Ramaligeswararao and Dr.V Sailaja and Dr.K. Srinivasa Rao},
  doi       = {10.14569/IJACSA.2011.021213},
  url       = {https://doi.org/10.14569/IJACSA.2011.021213}
}