Classification of audios containing speech and music

We propose an automated technique that uses perceptual and non-perceptual audio quality measures for discrimination of speech and music signals with high accuracy. Deployed audio quality measures used for characterization of audio are obtained via de-noising of the original audio. The underlying idea of the approach is that de-noising operation affects speech and music signals in a different and consistent manner and these differences can be captured by the audio quality metrics. Obtained quality measures are then used in conjunction with a machine learning classifier to statistically model speech and music signals. To determine the accuracy of the proposed method, tests have been performed on different datasets with and without audio compression.