RPMC-L2

The description is generated by Grok3.

Dataset download link: RPMC-L2

Dataset Structure

Data Instances

Each instance is an HDF5 file (.h5) containing synchronized music and lighting data for a specific live performance. The dataset is split into multiple parts (RPMC_L2_part_aa, RPMC_L2_part_ab, etc.) that can be merged into a single RPMC_L2.h5 file. Each file is organized into two main groups: music and light, with the following features:

• Music Features: Audio-related features stored as np.ndarray arrays with shape (X, L), where L is the sequence length.
• Light Features: Lighting-related data stored as np.ndarray arrays, primarily threshold data with shape (F, 3, 256).

Data Fields

Group	Feature	Shape	Description
music	openl3	(512, L)	OpenL3 deep audio embedding
music	mel_spectrogram	(128, L)	Mel spectrogram
music	mel_spectrogram_db	(128, L)	Mel spectrogram in decibels
music	cqt	(84, L)	Constant-Q transform (CQT)
music	stft	(1025, L)	Short-time Fourier transform (STFT)
music	mfcc	(128, L)	Mel-frequency cepstral coefficients
music	chroma_stft	(12, L)	Chroma features from STFT
music	chroma_cqt	(12, L)	Chroma features from CQT
music	chroma_cens	(12, L)	Chroma Energy Normalized Statistics
music	spectral_centroids	(1, L)	Spectral centroid
music	spectral_bandwidth	(1, L)	Spectral bandwidth
music	spectral_contrast	(7, L)	Spectral contrast
music	spectral_rolloff	(1, L)	Spectral rolloff frequency
music	zero_crossing_rate	(1, L)	Zero-crossing rate
light	threshold	(F, 3, 256)	Frame-specific light threshold data (Hue: 0–179, Saturation: 0–255, Value: 0–255)

Data Splits

The dataset consists of 699 files, organized by file hashes (top-level keys in the HDF5 file). There are no predefined splits; users can process the merged RPMC_L2.h5 file to create custom train/validation/test splits based on their research needs.

Dataset Creation

Curation Rationale

The dataset was created to facilitate research on the relationship between music characteristics and lighting effects in live performance venues, enabling applications in automated lighting design, audio-visual synchronization, and immersive live experiences.

Source Data

• Initial Data Collection:

  Data was collected from professional live performance venues hosting Rock, Punk, Metal, and Core music genres. Music features were extracted from audio recordings, and lighting data was captured as threshold values (Hue, Saturation, Value) synchronized with the audio.

  • Total Size: ~40 GB
  • Collection Method: Professional live performance venues
  • File Format: HDF5 (.h5)

• Who are the source data producers? The data was collected by researchers or professionals in live music venues, capturing synchronized audio and lighting data.

Annotations

• Annotation Process: The dataset includes music features (e.g., mel spectrogram, MFCC) and lighting data (threshold values for Hue, Saturation, Value) automatically extracted and synchronized during data collection. No manual annotations were provided.
• Who are the annotators? The dataset creators processed and organized the data using automated feature extraction tools.

Citation

@article{zhao2025automatic,
  title={Automatic Stage Lighting Control: Is it a Rule-Driven Process or Generative Task?},
  author={Zhao, Zijian and Jin, Dian and Zhou, Zijing and Zhang, Xiaoyu},
  journal={arXiv preprint arXiv:2506.01482},
  year={2025}
}