Free-text speaker identification over long distance telephone channel using hypothesized phonetic segmentation

摘要

Experimental investigation of a free-text speaker identification method based on long-term statistics was conducted on a widely used long-distance telephone database. On a 26-speaker subset, an average correct identification of 93.3% was obtained; on the complete 51-speaker set, 67.6% correct identification is obtained. Speaker verification experiments on the database provided receiver operating characteristics (ROCs) comparable to or better than the ones available in open literature.