Barking has been a controversial topic that has been studied from different points of view. While some authors argue that dog barking is a noncommunicative vocalization, others believe that barking plays a significant role in the human-dog interaction. Among the studies that take the last perspective, one of the most recent methods is to implement machine learning algorithms to classify single barks in different behavioral context by evaluating low-level descriptors. However, these research works do not incorporate the analysis of temporal structure or other dog vocalizations. In the present study, we proposed a broader approach by taking into account these relevant features that are currently not considered in the analysis of single barks for the classification of the context. By implementing an automatic process that segments long recordings of dog vocalizations and extracts both low-level and high-level descriptors, promising results were obtained for the barks' context classification from long recordings, where the highest value of F-measure was 0.71.