The 2005 AFRL/HEC One-Speaker Detection Systems
GENERAL DYNAMICS ADVANCED INFORMATION SYSTEMS DAYTON OH
Pagination or Media Count:
This paper describes the one-speaker detection systems submitted by AFRLHEC for several of the training and testing conditions in the 2005 NIST Speaker Recognition Evaluation. For each condition, the overall system score was the weighted combination of scores from several component systems. The component systems were based on mel-frequency cepstral coefficients MFCCs and Gaussian mixture models GMMs MFCCs and phoneme-specific GMMs PS-GMMs linear-prediction-based cepstral coefficients LPCCs from closed-phase analysis formant center frequencies, formant bandwidths, and fundamental frequencyFMBWFO and word language modeling WLM. The score combination was done using single-layer perceptrons, with the grouping of the component systems depending on the lengths of the training and testing files. For some of the testing andor training conditions involving ten-second speech files, the system performance improved from the inclusion of the FMBWFO and LPCC systems, while the MFCCPS-GMM system provided additional benefits in the one-conversation testing conditions involving larger amounts of training data.
- Numerical Mathematics
- Electricity and Magnetism
- Command, Control and Communications Systems