ROBUSTSPEAKERRECOGNITIONWITH CROSS-CHANNELDATA: MIT-LLRESULTSON THE 2006NISTSREAUXILIARYMICROPHONE TASK*

Oneparticularly difficult challenge forcross-channel speaker verification istheauxiliary microphone taskintroduced inthe 2005and2006NISTSpeaker Recognition Evaluations, where training usestelephone speech andverification usesspeech from multiple auxiliary microphones. Thispaperpresents two approaches tocompensate fortheeffects ofauxiliary microphones onthespeech signal. Thefirst compensation methodmitigates session effects through LatentFactorAnalysis (LFA)and Nuisance Attribute Projection (NAP).Thesecondapproach operates directly ontherecorded signal withnoisereduction techniques. Results arepresented thatshowareduction inthe performance gapbetween telephone andauxiliary microphone data. IndexTerms Speaker recognition, Speech enhancement, Microphones, Acoustic noise

[1]  Patrick Kenny,et al.  Eigenvoice modeling with sparse training data , 2005, IEEE Transactions on Speech and Audio Processing.

[2]  Douglas E. Sturim,et al.  SVM Based Speaker Verification using a GMM Supervector Kernel and NAP Variability Compensation , 2006, 2006 IEEE International Conference on Acoustics Speech and Signal Processing Proceedings.

[3]  Alvin F. Martin,et al.  The MMSR bilingual and crosschannel corpora for speaker recognition research and evaluation , 2004, Odyssey.

[4]  Patrick Kenny,et al.  Factor analysis simplified [speaker verification applications] , 2005, Proceedings. (ICASSP '05). IEEE International Conference on Acoustics, Speech, and Signal Processing, 2005..

[5]  Robert B. Dunn,et al.  Speech enhancement based on auditory spectral change , 2002, 2002 IEEE International Conference on Acoustics, Speech, and Signal Processing.