Automatic singing quality recognition employing artificial neural networks

The aim of the paper is to determine how quality of a singing voice can be recognized automatically. For this purpose, a database of singing voice sounds with samples of voices of trained and untrained singers was created and is presented. The methods of a singing voice parameterization are shortly reviewed and a set of descriptors is outlined. Each of the presented samples is parameterized and judged by experts, and the resulting feature vectors and quality scores are then used to train an artificial neural network. A comparison between experts' judgments and automatic recognition results is performed. Finally, statistical methods are applied to prove that an artificial neural network is able to automatically determine the quality of a singing voice with the accuracy very similar to expert assessments. The paper includes the discussion of results and presents derived conclusions.