Automatic Speaker Identification Using Nasal Spectra and Nasal Coarticulation as Acoustic Clues,
Abstract
Speaker identification was performed based upon the statistical information derived from the spectra of nasal consonants and words extracted from continuous speech as well as the spectra of continuous speech itself. Of the three clues, nasal consonants have been proved to be the best for speaker identification and words, least favorable. A new approach which uses the statistical properties of the nasal spectra was used to study the coarticulation of nasal consonants with vowels in isolated /h CVd/ utterances. The differences between the mean spectra of nasals followed by front vowels and followed by back vowels were used as the acoustic measure of the coarticulation of (m,n) and the following vowel context (V). The coarticulation between (n) and (V) was found to be about 30 percent of that between (m) and (V). Moreover, strong speaker idiosyncratic characteristics of the coarticulation process were found. Speaker identification was performed using the (m) coarticulation measure as an acoustic clue along with a correlation decision criterion. (Modified author abstract)
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1973
- Accession Number
- AD0773772
Entities
People
- King Sun Fu
- Lo-soun Su
Organizations
- Purdue University