Judgments of learning (JOLs) are sometimes influenced by factors that do not impact actual memory performance. One recent proposal is that perceptual fluency during encoding affects metamemory and is a basis of metacognitive illusions. In the present experiments, participants identified aurally presented words that contained inter-spliced silences (the generate condition) or that were intact, a manipulation analogous to visual generation manipulations. The generate condition produced lower perceptual fluency as assessed by both accuracy and identification latency. Consistent with the perceptual fluency hypothesis, the less fluent, generate condition produced lower JOLs than the intact condition. However, actual memory performance was greater in the generation than intact condition in free recall (Experiment 1) and recognition (Experiment 3). The negative effect of generation on JOLs occurred for both aggregate and item-by-item JOLs, but in the latter case, the positive generation effect in actual memory performance was reduced or eliminated (as also occurs with visual generation tasks; Experiments 2 and 4). Furthermore, the decrease in perceptual fluency produced by the generation manipulation was correlated with the decrease in JOLs for this condition (Experiment 5). The negative effect of generation on JOLs persisted even when participants were warned that the generation condition produces equal or greater memory performance compared to the intact condition (Experiment 6). The results are in accord with the perceptual fluency hypothesis and show that this metamemory illusion is related to objective measures of perceptual difficulty. With regard to actual memory performance, this novel auditory generation manipulation produces results consistent with those produced in the visual modality.