Feeling the Groove: Identification of Bass patterns from polyphonic audio

Technology has found its way into disparate spheres of music, from production to performance. Researchers have attempted to automate multitudinous aspects in this domain. One of the interests has been towards the automated generation of music pieces itself. Bass grooves are an integral part of most music pieces. It makes a piece sound complete and bridges the gap between the percussion and melody sections. Thus, it is essential for machines to understand bass grooves for automated music analysis and production. Automatically distinguishing bass grooves is difficult and it aggravates even more for polyphonic music. In polyphonic music, the bass grooves tend to be at a lower volume and its frequency range has profound overlap with the percussion section which contributes to the complexity of identification. In this paper, a system is presented to distinguish bass grooves in the presence of drums. Experiments were performed with 7 grooves totaling 4473 clips which were modeled using MFCC-based features. The highest accuracy of 97.38% was obtained using multi-layer perceptron (MLP)-based classification.