Criteria of voice characterization for speaker recognition
Abstract
A statistical model of voice recognition is studied. A covariance matrix of voice parameters was computed using data from 4 speakers, 20 samples and 28 parameters. Mahalanobis generalized distances were calculated using the covariance matrix. Each speaker is characterized by an average multidimensional vector. The maximum likelihood principle is applied to identify an unknown speaker by using the chi square test. The computed matrix data and individual voice parameters are presented.
- Publication:
-
NASA STI/Recon Technical Report N
- Pub Date:
- October 1982
- Bibcode:
- 1982STIN...8333039F
- Keywords:
-
- Mathematical Models;
- Multivariate Statistical Analysis;
- Speech Recognition;
- Voice Data Processing;
- Covariance;
- Matrices (Mathematics);
- Maximum Likelihood Estimates;
- Vectors (Mathematics);
- Communications and Radar