Honey bees (Apis spp.) are widely used as biological indicators of environmental changes. Recently, bees have been explored by researchers to monitor air contamination by listening to their hive sound. However, no study has determined whether beehive sound reflects the responses of bees to in-hive or out-hive chemicals. In this study, we conducted a feeding experiment to address this. First, we fed colonies with pure syrup (PS), syrup containing acetone (SA) or syrup containing ethyl acetate (SE) to collect beehive sound to establish multiple classifications using machine learning (ML) models. Then, we orderly fed colonies with PS, followed by SE, SA and PS. Next, we fed colonies in PS, SA, SE and PS order. Eventually, we evaluated the recall and precision of the model in detecting each syrup type. The result built on orderly feeding had a recall of 99%, 80%, 30%, 53% in detecting PS, SE, SA, PS, respectively. In the reverse feeding experiment, the ML model has a recall of 99%, 89%, 37% and 44% in detecting PS, SA, SE, and PS, respectively. Because the collected syrup in the two orderly feeding sessions was not removed from the frames during the experiment, the results indicate that beehive sound responds to chemicals in or out of the beehive.