On Speaker Identification,
Abstract
Automatic speaker identification was performed based on the statistical informaton extracted from power spectra of continuous speech. Nasal consonants extracted from the continuous speech have been proved to be the best features for speaker identification. Words without nasal consonants embeded are the least favorable. Both maximum correlation and minimum distance criteria are better than the minimum absolute difference criterion for speaker identification. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1972
- Accession Number
- AD0737751
Entities
People
- Lo-soun Su
Organizations
- Purdue University