MusER: Musical Element-Based Regularization for Generating Symbolic Music with Emotion

Authors: Shulei Ji, Xinyu Yang

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate that Mus ER outperforms the state-of-the-art models for generating emotional music in both objective and subjective evaluation. Experiments showed that Mus ER outperformed prior methods for generating emotional music in both objective music metrics and emotional expression.
Researcher Affiliation Academia Shulei Ji, Xinyu Yang School of Computer Science and Technology, Xi an Jiaotong University, Xi an 710049, P. R. China taylorji@stu.xjtu.edu.cn, yxyphd@mail.xjtu.edu.cn
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes We provide the generated music examples and code at a Git Hub repo: https://github.com/Tayjsl97/MusER.
Open Datasets Yes Dataset. Two datasets are used in this paper to train the proposed model. The first one is the Pop1k7 dataset1 (Hsiao et al. 2021), which contains 1748 piano covers of pop songs automatically transcribed by a piano transcription model (Hawthorne et al. 2018) and converted into MIDI files. The second one is the EMOPIA dataset2 (Hung et al. 2021), a multi-modal database focusing on perceived emotion in pop piano music. This dataset comprises 1087 music clips from 387 piano solo performances and clip-level emotion labels annotated by dedicated annotators.
Dataset Splits No The paper does not provide specific details regarding training, validation, and test dataset splits (e.g., percentages, sample counts, or citations to predefined splits).
Hardware Specification No The paper does not provide specific hardware details such as CPU/GPU models, memory specifications, or types of computing resources used for experiments.
Software Dependencies No The paper mentions using the 'Muspy library (Dong et al. 2020)' but does not specify version numbers for Muspy or any other key software components like Python, PyTorch, or TensorFlow.
Experiment Setup Yes Following (Hung et al. 2021), we set the length of the token sequence to 1024 for both datasets and apply specific sampling policies (temperature sampling and nucleus sampling (Holtzman et al. 2020)) for different elements.