"TalkPrinting": improving speaker recognition by modeling stylistic features

Authors:
Sachin Kajarekar;Kemal Sönmez;Luciana Ferrer;Venkata Gadde;Anand Venkataraman;Elizabeth Shriberg;Andreas Stolcke;Harry Bratt
Affiliations:
Speech Technology and Research Laboratory, SRI International, Menlo Park, CA;Speech Technology and Research Laboratory, SRI International, Menlo Park, CA;Speech Technology and Research Laboratory, SRI International, Menlo Park, CA;Speech Technology and Research Laboratory, SRI International, Menlo Park, CA;Speech Technology and Research Laboratory, SRI International, Menlo Park, CA;Speech Technology and Research Laboratory, SRI International, Menlo Park, CA;Speech Technology and Research Laboratory, SRI International, Menlo Park, CA;Speech Technology and Research Laboratory, SRI International, Menlo Park, CA
Venue:
ISI'03 Proceedings of the 1st NSF/NIJ conference on Intelligence and security informatics
Year:
2003

Citing 3
Cited 0

Introduction to statistical pattern recognition (2nd ed.)

Introduction to statistical pattern recognition (2nd ed.)
Speaker identification and verification using Gaussian mixture speaker models

Speech Communication
Prosody-based automatic segmentation of speech into sentences and topics

Speech Communication - Special issue on accessing information in spoken audio

Quantified Score

Hi-index	0.00

Visualization

Abstract

Automatic speaker recognition is an important technology forintelligence gathering, law enforcement, and audio mining. Conventionalspeaker recognition systems, which are based on independent short-term spectral samples, suffer from a lack of noise robustness and are unable to modela speaker's idiosyncratic stylistic features. This paper describes "TalkPrinting",a program of research aimed at adding such stylistic features to conventionalsystems. Results on three preliminary systems based on stylistic featuresdemonstrate that (1) the new features alone carry significant speakerinformation; (2) they also carry significant complementary informationcompared to the conventional features; and (3) they provide increasingimprovements in performance with increasing test durations.