B. Manaris, J. Romero, P. Machado, D. Krehbiel, T. Hirzel, W. Pharr, and R.B. Davis, "Zipf's Law, Music Classification and Aesthetics," Computer Music Journal 29(1), MIT Press, pp. 55-69, Spring 2005.
We present experimental results suggesting a connection between statistical proportions of music pieces and "pleasantness" (as reported by human subjects). We have created a collection of metrics based on Zipf's Law that measure the proportion of various parameters in music, such as pitch, duration, melodic intervals, and harmonic consonance. We applied these metrics to a large corpus of MIDI-encoded pieces. We used the generated data to perform statistical analyses and train artificial neural networks (ANNs) to perform various classification tasks, including "pleasantness" prediction. In particular, we trained an ANN using the data generated from a set of pieces together with human emotional responses to these pieces. Our hypothesis was that the ANN would discover correlations between Zipf-Mandelbrot distributions and human emotional responses. After training, the ANN was able to predict human aesthetic judgments of unknown pieces with very high accuracy.
This may have significant implications for music information retrieval and computer-aided music analysis and composition, and potentially it may provide insights on the connections among music, nature, and human physiology. We regard these results as preliminary; we hope they will encourage further investigation of Zipf's Law and its potential applications to music classification and aesthetics.
Index keywords: Zipf's law, aesthetics, classification, neural networks, quantitative analysis, author attribution, style identification, pleasantness prediction.