A voice conference system has a first voice-conference device having a speaker directed toward a first conferee and a second speaker directed toward a second conferee in one talk session room, and a second voice-conference device having a speaker directed to a third conferee and a fourth speaker directed toward a fourth conferee in another talk session room. When the first conferee talks, voice collecting beam signals picked up by microphones of the one voice conference device in excess of a predetermined threshold value are selected. The one voice conference device acquires voice collecting direction information, and then transmits voice communication data containing the voice collecting beam signal and the acquired voice collecting direction information to the another voice conference device, which acquires the voice collecting direction information from the received voice communication data, and emits the voice of the first conferee from the first speaker.