EUROSPEECH 2001 Scandinavia
In this paper we look at the parameterized feature-set that has been used in connected alpha-digit speech recognition and evaluate it on a speaker identification system. Compared to the popular mel-scaled feature-set (MFCC) the parameterized feature-set gives over 21% improvement in identification rate on the NTIMIT database in some cases. On average it shows a 14.0% improvement in identification rate. This demonstrates the improvement in performance that can be obtained using feature-sets.
Bibliographic reference. Mashao, Daniel J. / Baloyi, N. Tinyiko (2001): "Improvements in the speaker identification rate using feature-sets", In EUROSPEECH-2001, 2833-2836.