[Context and motivation] Conversations around requirements, such as interviews and workshops, are a key activity of requirements elicitation, and play a significant role in the creation of requirements specifications. [Question / problem] While these conversations contain a wealth of knowledge, requirements engineers use them mainly through note-taking during the conversation and by recalling the information from their memory. There is potential for supporting practitioners by retrieving important information from the recordings of these conversations. [Principal ideas / results] Although transcriptions can be automatically generated with good accuracy, they often contain too much text to be efficiently used for processing requirements elicitation sessions. Thus, we observed a need to transform these data sets into a useful format for requirements engineers to analyze. [Contribution] We present ReConSum, a prototype that utilizes natural language processing (NLP) to summarize requirements conversations. ReConSum takes as input a transcribed conversation, and it filters the speaker turns by keeping only those that include a question and that are expected to contain, or to be answered with, requirements-relevant information. In addition to presenting ReConSum, we experiment with different algorithms to assess the most effective combination.