On-Line Unsupervised Adaptation in Speaker Verification: Confidence-Based Updates and Improved Parameter Estimation

This paper presents the second part of a new approach to on-line unsupervised adaptation in speaker verification. The new approach extends previous work in the literature by (1) improving performance on the enrollment handset-type when adapting on a different handset-type (e.g., improving performance on cellular when adapting on a landline office phone), (2) accomplishing this cross channel improvement without increasing the size of the speaker model after adaptation, (3) employing a count-based, parameter-dependent smoothing algorithm that emphasizes the use of mean parameters in the speaker models until sufficient adaptation data are present to accurately estimate variances, and (4) developing a new confidence-based adaptation update weight which minimizes the corrupting effects on the speaker models from impostor attacks. Experimental results show a 61% (rel.) overall reduction in EER using the new on-line adaptation approach even with a significant impostor attack rate, and a 24% improvement in EER due to the new confidence-based adaptation scheme for those speaker models corrupted by impostor utterances.