During social interaction, animals integrate sensory information—including auditory cues—to influence future behavioral decisions. For example, animals respond to distinct vocalizations with unique actions (Seyfarth et al., 1980). Mice produce ultrasonic vocalizations (USVs) associated with discrete behaviors (Sangiamo et al., 2020); however, it is unclear if USVs inform subsequent actions. To investigate the predictive power of USVs on behavior, while mice freely interacted (n = 11 groups, 2 males and 2 females per group), we continuously recorded video and audio data using a sound source localization system with an 8-channel microphone array to accurately assign vocalizations to individuals. Using machine learning-based approaches to categorize vocalizations based on acoustic properties and extract behaviors in which two mice play unique roles (actor or recipient), we found a significant interaction between particular social behaviors and preceding vocalizations. A permutation test indicated that, depending on the behavioral role of the vocalizer, specific vocalizations preceded dominant (courtship or aggressive) or submissive (avoidant) behaviors at above-chance levels. Additionally, though classifiers trained to distinguish behavioral role based on preceding vocal emission failed for some behaviors, they were successful for particular dominant behaviors. These results suggest a potentially predictive relationship between acoustic communication and subsequent behaviors.