A Hilbert Space Approach to Linear Predictive Analysis of Speech Signals.
Abstract
Linear predictive analysis is geometrically interpreted to provide insight into the various formulations prevalent in current literature. The procedure for obtaining the predictive filter coefficients is considered as a minimum norm problem in an appropriate Hilbert space. Application of the projection theorem using specific sets of bases yields the normal equations for the covariance and autocorrelation methods. Orthogonalization of the basis vectors leads to the popular ladder structure and yields a recursive algorithm for evaluating the predictor and PARCOR coefficients. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1974
- Accession Number
- AD0776594
Entities
People
- Allen M. Peterson
- Kishan Shenoi
- Madihally J. Narasimha
Organizations
- Stanford University