This paper examines the influence of mismatch conditions on performance of a state-of-the-art system for forensic automatic speaker recognition (FASR) with known ground truth. The tested variables included generative AI output that is voice cloning, wearing a face mask, and increased physical effort. We hypothesized that these factors decrease the likelihood ratios when compared to baseline recordings. The outcomes of the comparison confirmed our assumptions for all hypotheses tested. The results showed a decrease in the likelihood ratios for testing the recordings with mismatch conditions when compared to the samples controlled for intrinsic and extrinsic mismatch. The findings of this study may appear relevant to automatic and semi-automatic speaker comparison and contribute to discussions on the application of automatic speaker identification methods to recordings with mismatch conditions.