We introduce the task of vocalist gender recognition in popular music and evaluate the benefit of Non-Negative Matrix Factorization based enhancement of melodic components to this aim. The underlying automatic separation of drum beats is described in detail, and the obtained significant gain by its use is verified in extensive test-runs on a novel database of 1.5 days of MP3 coded popular songs based on transcriptions of the Karaoke-game UltraStar. As classifiers serve Support Vector Machines and Hidden Naive Bayes. Overall, the suggested methods lead to fully automatic recognition of the pre-dominant vocalist gender at 87.31% accuracy on song level for artists unkown to the system in originally recorded music.