|
| |
|
|
Controlled Music Generation with Deep Learning
Židek, Marek ; Hajič, Jan (advisor) ; Matzner, Filip (referee)
Generation of musical compositions is one of the hardest tasks for artificial intelligence where most of the current approaches struggle with long term coherence of the generated compositions. This work aims to demonstrate how deep learning models for generating music can be externally controlled to produce compositions with long term coherence, polyphony, and multiple instruments. We work with classical music ranging from compositions for piano through string quartet and up to symphonic orchestral compositions. To control the generation process, we take inspiration from the abstract notion of musical form: normally a high-level description of how similar and dissimilar passages are arranged throughout a composition, we use it as a recipe for generating a coherent composition. To this end, we (1) design a sufficiently general music similarity pseudometric from existing methods, (2) extract musical form from the training data by applying a clustering algorithm over the similarity values, (3) train three models that generate similar and locally coherent dissimilar musical fragments, and (4) design a way how to use the musical forms during the generation process to orchestrate the inference of the three models to generate whole compositions from musical fragments. We show what is the performance of the...
|
|
|
Generating polyphonic music using neural networks
Židek, Marek ; Hajič, Jan (advisor) ; Maršík, Ladislav (referee)
The aim of this thesis is to explore new ways of generating unique polyphonic music using neural networks. Music generation, either in raw audio waveforms or discretely represented, is very interesting and under a heavy ex- ploration in recent years. This thesis works with midi represented polyphonic classical music for piano as training data. We introduce the problem, show rele- vant neural network architectures and describe our numerous ideas, out of which one idea, our experiment with three versions of skip residual LSTM connections for music composition, we consider a good contribution to the field. In related work, skip-connections were explored mostly for classification tasks, however, our results show a solid improvement for music composition (e.g. 47% of respondents considered our samples real). We also show that skip-connections have rather diverse hyperparameter space for future tuning. Apart from standard automated test set evaluation, which is hard to design and interpret for creativity mimicking models, we also did a complex evaluation through surveys. The evaluation was specifically designed to not only to show results for our samples, but to reveal information about expectancy, preconceptions and influence of personal charac- teristics of the respondents. We consider this a valuable...
|
|
| |
guest ::
