A method and an audio conference device for carrying out an audio conference are disclosed, whereby classification information associated with a respective audio date flow is recorded for supplied audio data flows. According to a result of an evaluation of the classification information, the audio data flows are associated with at least three groups which are homogeneous with regard to the results. The individual audio data flows are processed uniformly in each group in terms of the signals thereof, and said audio data flows processed in this way are superimposed in order to form audio conference data flows to be transmitted to the communication terminals. Further, an audio conference device and a method for switching between a first encoder and a second encoder for an audio data connection are disclosed, wherein the encoders use, inter alia, encoding parameters influenced by an audio data history, for the encoding process. The audio data connection from the first encoder to the second encoder is switched in such a way that, during the switching process, the encoding parameters of the second encoder are matched with the encoding parameters of the first encoder, and when the encoding parameters are matched, the audio connection is switched to the second encoder.