Automatic Music Composition with Simple Probabilistic Generative Grammars
Horacio Alberto García Salas, Alexander F. Gelbukh, Hiram Calvo, Fernando Galindo Soria · Polibits · 2011
"We propose a model to generate music following alinguistic approach. Musical melodies form the training corpuswhere each of them is considered a phrase of a language.Implementing an unsupervised technique we infer a grammar ofthis language. We do not use predefined rules. Music generationis based on music knowledge represented by probabilisticmatrices, which we call evolutionary matrices because they arechanging constantly, even while they are generating newcompositions. We show that the information coded by thesematrices can be represented at any time by a probabilisticgrammar; however we keep the representation of matricesbecause they are easier to update, while it is possible to keepseparated matrices for generation of different elements ofexpressivity such as velocity, changes of rhythm, or timbre,adding several elements of expressiveness to the automaticallygenerated compositions. We present the melodies generated byour model to a group of subjects and they ranked ourcompositions among and sometimes above human composedmelodies."