Speaker verification with long-term ageing data

Abstract

Abstract

Speaker verification in score-ageing-quality classification space

“…Along with classical quality measures, we propose and evaluate a new model-based quality measure. We demonstrate the effectiveness of this measure at predicting score degradation due to quality variation, and use it to improve the performance of our long-term ageing system, first presented in Kelly et al (2012).…”

“…In this paper we extend the approach in (Kelly et al, 2012) by incorporating objective measures of recording quality. Firstly, this allows us to reduce non-ageing-related variability and further isolate the ageing effect.…”

“…In our recent work (Kelly et al, 2012), the Trinity College Dublin Speaker Ageing (TCDSA) database, containing 18 speakers with recordings spanning 30-60 years per speaker, was presented. A speaker verification evaluation of the database demonstrated a progressive degradation in verification score with ageing.…”

Multi-way regression for age prediction exploiting speech and face image information

“…The proposed age estimation framework is initially applied to the longitudinal TCDSA database [13] for speech-based age prediction. The database contains recordings spanning a year range per speaker varying between 30 and 60 years at irregular intervals between 1 to 10 years.…”

“…Experiments are conducted for age estimation based solely on speech, solely on face images, and on both speech utterances and face images. To this end, the Trinity College Dublin Speaker Ageing database (TCDSA) [13] is supplemented with face images of the speakers, which are contemporary to their speech recordings. Moreover, experiments are conducted on the benchmark FG-NET Aging dataset in order to illustrate better the performance of the proposed age estimation framework.…”