In this paper, we investigate the differences betweenhuman-directed speech and robot-directed speech duringspontaneous human-human-robot interactions. Theinteractions under study are different from previousstudies, in the sense that the robot has a more similarrole as the human interlocutors, which leads to morespontaneous turn-taking. 20 conversations were extractedfrom a multi-party human-robot discussioncorpus, where two humans are playing a collaborativecard game with a social robot. Each utterance inthe conversations was manually labeled according toaddressee (robot or human). The following acousticfeatures were extracted: fundamental frequency, intensity,speaking rate, and total utterance duration.There were significant differences between humanandrobot-directed speech for speaking rate and thetotal utterance duration. These results are in linewith previous studies on robot-directed speech, andconfirms that this difference holds also when the conversationsare of a more spontaneous nature.