Download README.md from ben2002chou/notesep-training-data: direct link, hf CLI and curl.
- Browser
- Download file 2.27 kB
-
https://huggingface.co/datasets/ben2002chou/notesep-training-data/resolve/main/README.md
- Command line
-
hf download hf://datasets/ben2002chou/notesep-training-data/README.md
-
curl -L -o README.md https://huggingface.co/datasets/ben2002chou/notesep-training-data/resolve/main/README.md
pretty_name: NoteSep NoteGrab training data
license: other
license_name: notesep-source-licenses
license_link: >-
https://huggingface.co/datasets/ben2002chou/notesep-training-data/blob/main/SOURCES.md
tags:
- music
- audio
- source-separation
NoteGrab training data
Two million score-addressed note targets from 100,696 synthetic mixtures, plus eight recorded-scene shards. A mixture may be reused for several target notes. Each index row identifies the mixture, requested note, and waveform data.
| Path | Contents |
|---|---|
| synthetic/ | Synthetic audio, training-row index, timing and sampling metadata. |
| recorded/ | Recorded scenes, isolated note targets and their indexes. |
| example/ | One unpacked mixture, isolated target, reference sound and note labels. |
| manifest.json | File inventory, compression information and training path mappings. |
Download and restore the package from the NoteSep repository:
uv run notesep download-data --repo-id ben2002chou/notesep-training-data --output data/notegrab
Setup verifies the files, decompresses audio shards, and writes path-map.json for training. The compressed download is about 446 GB. Restored data occupies about 2.39 TB; allow space for both during setup. Audio values are preserved exactly. The training guide in the code repository gives the training command.
Start with example/ to listen to a training row. The larger files are indexed containers for efficient training, rather than directories of individual WAVs. ASO uses separate source-complete scenes and locally generated NoteGrab estimates.
synthetic/sampling-parent.json selects the rows for the original 30,000-update
parent run. synthetic/sampling.json selects rows for the later continuation.
The intermediate 250-update stage uses synthetic/sampling-adaptation.json
and the instrument-library labels in synthetic/render-methods.npz. All three
row selections refer to the same audio and row index.
Sources and licenses
See SOURCES.md for instrument, score and recording credits, transformations and source-specific terms. Original notices are included in licenses/ and source information in sources/. Noncommercial and share-alike terms apply to the corresponding source material.