Speaker discrimination using short- and long-term segmental information in vowels

Abstract

Apart from carrying linguistic information, vowels have been shown to provide speaker-specific cues. We compared the contribution to the discrimination between speakers of both static and dynamic formant parameters.

The overall classification rate for all parameters reaches 46.5 %. Spectral slope characteristics were outperformed by formant means and those, in turn, outperformed by dynamic formant trajectories.

Keywords

• disriminant analysis
• formant means
• formant trajectories
• speaker identification