Speaker Identification Using the Two-Dimensional Cepstrum Transform.
Abstract
This thesis investigates the application of the two-dimensional cepstrum transform to a speaker identification System. Two distance measures are implemented for identification decision; the Euclidean distance and a weighted two-dimensional cepstral distance. The study considers three words to be tested under several noise levels. The effect of speaking rate during recordings is examined and is shown to be critical. Results show identification rates in the range of 95% to 98.5% for 50 dB signal to noise ratio and 57.65% to 80.7% for 0 dB signal to noise ratio.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1995
- Accession Number
- ADA294204
Entities
People
- Ioannis Lelakis
Organizations
- Naval Postgraduate School