Open Music Datasets For Generative AI
WaivOps is a crowd-sourced music creator project focused on providing newly sourced music datasets for developing generative AI models. Unlike commonly sourced audio and MIDI training data, these datasets are specifically developed to enhance model performance by augmenting and generalizing musical notations. Each dataset is developed with model objectives in mind, offering developers and AI music enthusiasts access to large-scale data for various tasks involved in pretraining and fine-tuning models.
The datasets are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). For more detailed information, please visit the links provided to GitHub or Zenodo.
POP-ROK Dataset
POP-ROK is an open audio dataset consisting of 5,378 synthetic drum recordings in the style of pop rock music, provided in uncompressed stereo WAV format along with paired JSON files. The dataset was developed by sonifying approximately 30 acoustic drum kits with MIDI and employs various data augmentation techniques to create diverse drum kits and acoustic environments.
Size: 12.6 GB
SYN-SE1 Dataset
SYN-SE1 is an open audio dataset containing archived recordings of a Studio Electronics SE1 analog synthesizer. It includes 1,000 one-shot audio samples recorded in uncompressed stereo WAV format, labeled by note key across a two-octave range. The presets encompass a variety of distinct synth bass and lower-pitched lead sounds, featuring filter modulations and spatial stereo imaging, providing a valuable resource for soundfont design, audio production, and training data for generative AI models.
Size: 75.6 MB
WRLD-SMB Dataset
EDM-HSE is an open audio dataset containing a collection of code-generated drum recordings in the style of modern electronic house music. It includes 8,000 audio loops recorded in uncompressed stereo WAV format, created using custom audio samples and a MIDI drum dataset. The dataset also comes with paired JSON files containing MIDI note numbers (pitch) and tempo data, intended for supervised training of generative AI audio models.
Size: 3.1 GB
EDM-HSE Dataset
EDM-HSE is an open audio dataset containing a collection of code-generated drum recordings in the style of modern electronic house music. It includes 8,000 audio loops recorded in uncompressed stereo WAV format, created using custom audio samples and a MIDI drum dataset. The dataset also comes with paired JSON files containing MIDI note numbers (pitch) and tempo data, intended for supervised training of generative AI audio models.
Size: 7.6 GB
RGTM-PNO Dataset
RGTM-PNO is an open audio dataset featuring a collection of vintage piano songs in the style of ragtime, a genre that flourished around the turn of the 20th century. The dataset contains 262 audio tracks recorded in uncompressed stereo WAV format, synthetically generated using a custom soundfont and MIDI files sourced from public resources online.
Size: 7.5 GB
EDM-TR8 Dataset
EDM-TR8 is an open audio dataset composed of a series of drum recordings in the style of electronic dance music (EDM). This dataset primarily focuses on the iconic sounds of the Roland TR-808 drum machine with additional electro synth drums. The dataset contains 3,790 audio loops recorded in uncompressed stereo WAV format, generated with custom audio samples and a MIDI dataset used for training symbolic music models.
Size: 4.4 GB
EDM-TR9 Dataset
EDM-TR9 is an open audio dataset composed of a series of drum recordings in the style of electronic dance music (EDM). This dataset primarily focuses on the distinctive sounds and rhythm patterns of the Roland TR-909 drum machine within the subgenres of dance, house and techno music. The dataset contains 3,780 audio loops recorded in uncompressed stereo WAV format, produced with custom drum samples and MIDI-programmed rhythms at various tempo rates.
Size: 4.8 GB
RTRO-DRM Dataset
RTRO-DRM is an open audio dataset composed of a series of drum recordings in the style of 1980s electronic music. The dataset comprises 2,138 raw, unedited audio clips recorded in uncompressed stereo WAV format. These recordings were curated using an internal drum sample dataset and MIDI files sourced from a code-based music generation system, along with a MIDI transformer model trained on more than 30,000 MIDI files. The files primarily consist of recordings that may not meet conventional audio quality standards but can still be valuable for a range of applications and research projects.
Size: 3.7 GB
WRLD-LP Dataset
WRLD-LP is an open audio dataset comprised of a series of symbolic drum recordings in the genres of world percussion music. The dataset includes 3,162 audio loops recorded in uncompressed stereo WAV format. The compositions were generated with an internal sample dataset played with note-dense MIDI drum files from a code-based music generation system, along with a MIDI transformer model trained on more than 30,000 MIDI files.
Size: 6.5 GB
HH-LFBB Dataset
HH-LFBB is an open audio dataset composed of a series of drum recordings in the style of lofi hip-hop music. The dataset contains 3,332 audio loops recorded in uncompressed stereo WAV format, produced with custom drum samples and MIDI-programmed rhythms at various tempo rates.
Size: 15.1 GB