A Factored Language Model of Quantized Pitch and Duration

This paper investigates a novel statistical approach to mu- sic classication that utilizes recent technology developed in the domain of natural language processing. Specically , we investigate the use of factored language models (FLMs) for the task of producing conditional probability distributions to model origin-specic folk songs. In our model, pitch cluster and quantized duration are employed as the two fundamental factors in a musical unit. The structure of our FLM is empir- ically chosen from data to minimize perplexity. We apply this collection of FLMs to the task of folk song classication. Our experiments show classication accuracy of 72.5% on a data set of European folk songs from 6 different regions.