Generování žánrově specifické hudební transkripce Antonína Dvořáka prostřednictvím variačního autoenkodéru

Abstract

Apart from traditional deep learning tasks such as pattern recognition, stock price prediction, and machine translation, this method also finds practical application within algorithmic composition. This paper explores the use of a generative model based on unsupervised learning of a musical style from a pre-selected corpus and the subsequent prediction of samples from the estimated distribution. The model uses a Long Short-Term Memory neural network whose training data contains genre-specific melodies in symbolic representation.

Keywords

• algorithmic composition
• artificial intelligence
• autoencoder
• deep learning
• generative art
• LSTM network