In this talk I plan to present my most recent directions of exploration in the area of speaker recognition. A description of the keyword HMM system submitted for the NIST 2004 Speaker Recognition Evaluation will be given along with subsequent modifications aimed at enhancing system performance. As a reminder, this system seeks to perform text-constrained speaker recognition in text-independent domains by limiting the words of interest to those occurring with high frequency in the domain. I will also present and give preliminary results for a contrastive system using monophone HMMs--an attempt to explore the trade-off between token coverage and "sharpness" of modeling. Finally, I will give mention of a hybrid SVM/HMM system using features derived from keyword HMM parameters.