A method and system are provided. The method includes transcribing into text, by a processor-based voice-to-text transcriber, voices uttered by two or more respective participants other than a user in each of one or more voice communications sessions occurring on multiple communication devices. The method further includes generating, for each of the one or more voice communication sessions, a respective prediction from the text of when a topic of interest to the user is likely to be brought up therein. The method also includes connecting the user to at least one of the one or more voice communications sessions, responsive to the respective prediction therefor such that the user participates in only a subset of an overall conversation in the at least one of the one or more voice communication sessions.