MoisesDB

1. Download the dataset from the Moises research website.
2. Extract the files.
3. Set the MOISESDB_PATH environment variable to point to the extracted directory.

The directory structure should follow this pattern:

moisesdb:
    moisesdb_v0.1
        track uuid 0
        track uuid 1
        ...

export MOISESDB_PATH=./moises-db-data

You can install the moises-db package directly from the GitHub repository using pip.

pip install git+https://github.com/moises-ai/moises-db.git

To ensure your downloaded moisesdb.zip is not corrupted, verify its hashes against the following values:

MoisesDB Hashes:

• MD5: 13cf74eda129c38b914a51ea79fb1778
• SHA256: 4cde33ce416ac7c868cffcb60eb31f5c741ab7ae5601cbb9d99ed498b72c48c1

Verification Commands:

Linux/Mac:

md5sum moisesdb.zip
sha256sum moisesdb.zip

Windows (PowerShell):

Get-FileHash -Algorithm MD5 moisesdb.zip
Get-FileHash -Algorithm SHA256 moisesdb.zip

A MoisesDBTrack object represents a single track and provides access to its stems, mixture, and metadata. Stems and mixtures are computed on-the-fly.

Audio Properties

• track.stems: A dictionary where keys are stem names (e.g., 'vocals', 'bass') and values are nd.array objects.
• track.audio: An nd.array representing the full mixture.

Metadata Properties

• track.id
• track.provider
• track.artist
• track.name
• track.genre
• track.sources
• track.bleedings
• track.activity

Saving Stems

You can persist the on-the-fly computed stems to disk using the save_stems method.

track = db[0]

# Access audio
stems = track.stems  # stems = {'vocals': ..., 'bass': ..., ...}
mix = track.audio    # mixture = nd.array

# Save stems to a specific directory
path = './moises-db-stems/0'
track.save_stems(path)

Use the MoisesDB class to access tracks in the dataset. You can either rely on the MOISESDB_PATH environment variable or explicitly provide the data_path argument. You can also specify the sample_rate.

The MoisesDB object supports standard Python indexing and length operations.

from moisesdb.dataset import MoisesDB

db = MoisesDB(
    data_path='./moisesdb',
    sample_rate=44100
)

n_songs = len(db)
track = db[0]  # Returns a MoisesDBTrack object