This study focused on analyzing how different interactions (face-to-face chatting and chatting with AI chatbots) may affect speaking skills. In order to examine whether using AI chatbots can be beneficial for improving speaking task performances, 110 participants were divided into three groups: a face-toface chatting, AI text-chatting, and AI voice-chatting group. They were assigned three speaking tasks that involved describing a picture, responding to questions, and expressing an opinion, which they did both before and after the experiments. Based on the analysis of quantitative data, the findings of the study indicated that both AI groups improved their speaking performance after the experiment while there were mixed results from the face-to-face interactions. It was found that the participants in the face-to-face group improved significantly better in responding to questions and expressing an opinion while no significant differences were revealed in the describing a picture task. With respect to the effects of using AI chatbots, there were no significant differences in the two speaking tasks (describe a picture and respond to questions) between the three groups. However, the comparative analysis of the three different interaction modes suggested that the AI voice-chatting pairs outperformed in terms of their speaking performance task (expressing an opinion) compared with the face-to-face and AI textchatting groups. Regarding students' perspectives, descriptive statistics and selected student interviews were reported. Based on the findings of the current study, pedagogical implications and directions for future research are also suggested.