Vocal communication is integral to the social behavior of vertebrates, with birds and mammals exhibiting especially sophisticated vocal interactions. Investigating these interactions and their neural underpinnings requires experimental designs that minimally disrupt natural behaviors. In this context, fully automated experimental setups are highly advantageous. Bats, known for their complex vocal repertoire and learning capabilities, engage in rapid vocal exchanges, posing a challenge for real-time analysis and interaction. Our research aimed to develop an automated MATLAB® based tool that leverages Deep Learning Toolbox™ and Audio Toolbox™ to classify bat communication sounds for immediate playback during behavioral and neurophysiological studies with free-ranging bats. We trained a neural network to classify bat calls into four behaviorally relevant types such as appeasement and aggression calls. Recordings were captured using an ultrasonic microphone and digitized at 192 kHz. The network, pretrained on the four call categories, automatically detected and classified individual calls. Depending on the classification, a corresponding sound was played back, mimicking behaviorally relevant response times and enabling interactive communication with the bats. The system also incorporated out-of-distribution detection to differentiate calls from noise.Our results showcase a classification accuracy of over 93% on test data and a behaviorally relevant response time, mimicking natural communication, and highlighting the tool's efficacy for real-time interactions. This method offers a rapid and user-friendly solution for probing vocal communication mechanisms across various animal models, making it a significant advancement for experimental studies in the field.