Consistent confusions --- word misperceptions reported in an open set task with a high agreement across listeners --- can be especially valuable in understanding the detailed processes underlying speech perception. The current study investigates the origin of a set of consistent confusions collected in a variety of masking conditions, by applying signal-level modifications to the stimuli eliciting the confusion, and subsequently reevaluating listeners' percepts. Modifications were selected to provide release from either the energetic or the informational component of the maskers and involved manipulations of signal-to-noise ratio, fundamental frequency, and resynthesis of the noise-mixture in glimpsed regions of the target speech. Increasing signal-to-noise ratio and glimpse resynthesis showed the expected release from energetic and informational masking respectively. However, manipulations targeting informational masking release, including fundamental frequency modification, affected a surprisingly high number of confusions stemming from energetic maskers. The degree of fundamental frequency shift did not have a significant effect on the response patterns observed. Around 30\% of confusions can be explained solely based on the information contained within the target glimpses surviving energetic masking, while for the rest of the cases additional factors, such as recruitment of information from the masker, appear to be involved.