Cognitive models of memory retrieval have been applied in learning systems that aid in memorizing information by adapting to the needs of individual learners. More specifically, such systems present retrieval questions of appropriate difficulty to learners to help them memorize factual information like vocabulary or glossary items. To estimate appropriate item repetition schedules, traditional learning systems use accuracy and response times as behavioral proxies of memory strength. Here, we focus on spoken responses to visually presented retrieval cues. In two experiments, we analyzed over 10,000 utterances and demonstrate that it is possible to extract information about (a) the accuracy of a response, and (b) a speaker’s subjective confidence in a response, from prosodic cues in the speech signal. Furthermore, we show that a learner’s meta-cognitive beliefs about memory performance are reflected mainly by variations in pitch and speaking speed, whereas the objective accuracy of the response is mainly related to its loudness. Fundamentally, these results contribute to a better understanding of the relationship between prosodic speech variations and (meta)memory processes, and could facilitate the development of speech analyses as a tool in learning research. Practically, they can improve models of memory retrieval used in adaptive learning systems. For example, extracting information about a speaker’s confidence from the speech signal in real time may allow for improvement of predictions of future retrieval success—without the learner having to make explicit confidence judgments after each learning trial.