A timing-based classification method for human voice in opera recordings

The goal of this work is to identify famous tenors from commercial recordings. Our approach is based on training expressive singer-specific models and using them to classify new musical fragments interpreted by singers that perform arias from the training set. In this paper we focus on expressive timing variations and build the models by applying machine learning techniques to a body of data consisting of high-level descriptors extracted from audio recordings. The experimental results show evidence that performers can be automatically identified at a rate significantly better than random choice.