Interspeech'2005 - Eurospeech
This paper investigates the task of speaker verification in noisy conditions, assuming no prior knowledge about the noise. To achieve this we use subband features which can isolate band-limited corruption and allow verification to be performed on mainly clean subbands. To further improve speaker verification performance, we propose a novel method to model correlation between subband feature streams. In this method, all possible combinations of subbands are created and each combination is treated as a single frequency band by calculating a single feature vector for it. The resulting feature vectors capture information about every band in the combination as well as the dependency across the bands. Experiments conducted on the NIST 1998 database demonstrate improved robustness for the new model, in the presence of both stationary and non-stationary noise.
Bibliographic reference. McAuley, James / Ming, Ji / Corr, Pat (2005): "Speaker verification in noisy conditions using correlated subband features", In INTERSPEECH-2005, 2001-2004.