A system of microphones, signal processors, and loudspeakers provides enhanced comprehension of speech in noisy social events where the locations of participants are relatively constrained. Speech enhancement comprises both boosting sounds moderately at higher frequencies and delaying them to match the arrival of sounds directly from speakers. Virtual loudspeakers create the illusion for each listener that the boosted sounds come from the vicinity of each talker. Video cameras determine the head positions and facing directions of participants. Subgroups of participants having at least temporary conversational affinities can be identified. Such subgroups may overlap and change. Speech from talking participants is picked up by directional microphones and filtered, sorted, and relayed selectively via loudspeakers to listening participants identified as subgroup members, reinforcing and enhancing the naturally heard speech. More weight can be given to enhancing speech between affined participants. Either conventional or parametric loudspeakers may be used.