Empirical Mode Decomposition Applied to Speaker Identification

Open Access
Bartuska, Kevin Michael
Area of Honors:
Electrical Engineering
Bachelor of Science
Document Type:
Thesis Supervisors:
  • John Doherty, Thesis Supervisor
  • Jeffrey Scott Mayer, Honors Advisor
  • Signal Processing
  • Empirical Mode Decomposition
  • Speaker Identification
  • Speaker Recognition
This project applied the algorithm “Empirical Mode Decomposition” (EMD) to the problem of speaker identification – that is, recognizing a speaker by their unique tone of voice. A Gaussian mixture model was used to compare the mel frequency cepstral coefficients (MFCCs) of the speech files, after their processing with EMD. This experimenter determined that only the first several EMD modes contained the relevant information for speaker identification. Additionally, EMD was combined with singular value decomposition, but the benefits to recognition rate were not significant.