Open ZFTurbo opened 1 month ago
Dataset name: MUSDB18-HQ Description: Most popular dataset for Music Source Separation Instuments: bass, drums, vocals, other Format: Sample rate 44100 Hz, wav. Volume: 100 train songs, 50 test songs Size: 22.7GB Download link: https://zenodo.org/records/3338373
Dataset name: StemGMD Description: A Large-Scale Audio Dataset of Isolated Drum Stems for Deep Drums Demixing Instuments: Drums [Kick Drum, Snare, High Tom, Low-Mid Tom, High Floor Tom, Closed Hi-Hat, Open Hi-Hat, Crash Cymbal, Ride Cymbal]
Format: 16-bit/44.1 kHz stereo WAV file using ten realistic-sounding acoustic drum kits sourced from the Logic Pro X sample libraries
Volume: StemGMD contains 1224 hours of audio, which correspond to more than 136 hours of full-kit mixtures Size: 1.13 TB Download link: https://zenodo.org/records/7860223 and https://zenodo.org/records/7882857
Thoughts : Even if it was made simultaneously than LarsNet by same authors, it's maybe not ideal for source separation as it's really lacking sound diversity (only 10 differents drumkits), but elements can be mixed with other kits elements (and from a sound mixing engineer point of vue, it could also have been way better).
Dataset name: MoisesDB Description: MoisesDB is a comprehensive multitrack dataset for source separation beyond 4-stems, comprising 240 previously unreleased songs by 47 artists spanning twelve high-level genres. The total duration of the dataset is 14 hours, 24 minutes and 46 seconds, with an average recording length of 3:36 seconds. MoisesDB is offered free of charge for non-commercial research use only and includes baseline performance results for two publicly available source separation methods. Instuments: Bass, bowed strings, drums, guitar, other, other keys, other plucked, percussion, piano, vocals, wind Format: I didn't download it yet but it seems to be 44100 Hz wav Volume: 240 songs in 14 hours 24 minutes and 46 seconds Size: 82.7 GB Download link: https://developer.moises.ai/research, https://github.com/moises-ai/moises-db
Here you can find more (but not sorted as neat as in the form above): https://docs.google.com/document/d/17fjNvJzj8ZGSer7c7OFe_CNfUKbAxEh_OBv94ZdRG5c/edit#heading=h.k3cm3bvgsf4j
Dataset name: The MAESTRO v3.0.0 Description: MAESTRO (MIDI and Audio Edited for Synchronous TRacks and Organization) is a dataset composed of about 200 hours of virtuosic piano performances captured with fine alignment (~3 ms) between note labels and audio waveforms. Instuments: piano Format: Sample rate 44100 Hz, wav. Volume: 1286 songs (~200 hours) Size: 120 GB Download link: https://magenta.tensorflow.org/datasets/maestro#v300
We can try to gather all public and private datasets in one place. To post dataset, please, fill the form: