15th International Congress of Phonetic Sciences (ICPhS-15)

Barcelona, Spain
August 3-9, 2003

Development of a Multiparametric Speaker Profile for Speaker Recognition

Antti Iivonen (1), Kirsi Harinen (1), Leena Keinanen (1), Jussi Kirjavainen (1), Einar Meister (2), Launo Tuuri (1)

(1) University of Helsinki, Finland
(2) Tallinn Technical University, Estonia

This report includes (1) recording and storing of different Finnish speech materials including a mobile telephone (GSM) speech data base (about 240 speakers), (2) critical composition of an acoustical multiparametric speaker profile, (3) development of a speaker identification program PROFMATCH based on the use of individual profiles, and (4) clustering and identification tests with hi-fi, medium and GSM quality speech. An automatic multiparametric Praat script series SpeakerProfiler has been created. An identification score 100% was achieved with 110 speakers in the tests with medium technical speech quality utilising Gaussian Mixture Model and only few basic parameters. Identification of GSM quality was successful, if the recording channel was unchanged. More tests are needed for an improvement of the recognition score, if variable recording channels are applied. Attempts with an automatic segmentation program (SEGMENTER) yielding an access to individual sound segments have been promising.

Full Paper

Bibliographic reference.  Iivonen, Antti / Harinen, Kirsi / Keinanen, Leena / Kirjavainen, Jussi / Meister, Einar / Tuuri, Launo (2003): "Development of a multiparametric speaker profile for speaker recognition", In ICPhS-15, 695-698.