The application of community detection in complex networks is explored within the framework of cover song identification, i.e. the automatic detection of different audio renditions of the same underlying musical piece. In the last years this particular task has been widely studied within the music information retrieval field as a query problem, where one song was submitted and a list of possible matches was created by the system. In this contribution we propose a new point of view: songs are embedded in a complex weighted network, whose links represent similarity (common musical content between songs). We analyze this network and find a strong modular structure, with well-defined communities and a clustering coefficient higher than expected. We then perform clustering and community detection to identify groups of songs that are versions the same musical piece. Importantly, the information gained through this process can be used to increase the overall accuracy of the system. Results show that accuracy increments of 5 percent points can be easily achieved. A further out-of-sample test provides evidence that this increase can be potentially higher.