Self-supervised pretraining on 60,000 hours of symbolic piano music produces a generative model and contrastive embeddings that beat leading baselines on continuation quality and several MIR classification benchmarks.
Learning to Generate Music With Sentiment
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
abstract
Deep Learning models have shown very promising results in automatically composing polyphonic music pieces. However, it is very hard to control such models in order to guide the compositions towards a desired goal. We are interested in controlling a model to automatically generate music with a given sentiment. This paper presents a generative Deep Learning model that can be directed to compose music with a given sentiment. Besides music generation, the same model can be used for sentiment analysis of symbolic music. We evaluate the accuracy of the model in classifying sentiment of symbolic music using a new dataset of video game soundtracks. Results show that our model is able to obtain good prediction accuracy. A user study shows that human subjects agreed that the generated music has the intended sentiment, however negative pieces can be ambiguous.
fields
cs.SD 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Scaling Self-Supervised Representation Learning for Symbolic Piano Performance
Self-supervised pretraining on 60,000 hours of symbolic piano music produces a generative model and contrastive embeddings that beat leading baselines on continuation quality and several MIR classification benchmarks.