Welcome to musdb documentation!

musdb

musdb.audio_classes

musdb.tools

API documentation

class musdb.DB(root=None, setup_file=None, is_wav=False, download=False, subsets=['train', 'test'], split=None, sample_rate=None)[source]

The musdb DB Object

Parameters
  • root (str, optional) – musdb Root path. If set to None it will be read from the MUSDB_PATH environment variable

  • subsets (list[str], optional) – select a _musdb_ subset train or test (defaults to both)

  • is_wav (boolean, optional) – expect subfolder with wav files for each source instead stems, defaults to False

  • download (boolean, optional) – download sample version of MUSDB18 which includes 7s excerpts, defaults to False

  • subsets – select a _musdb_ subset train or test. Default None loads [‘train’, ‘test’]

  • split (str, optional) – when subsets=train, split selects the train/validation split. split=’train’ loads the training split, `split=’valid’ loads the validation split. split=None applies no splitting.

setup_file

path to yaml file. default: setup.yaml

Type

str

root

musdb Root path. Default is MUSDB_PATH. In combination with download, this path will set the download destination and set to ‘~/musdb/’ by default.

Type

str

sources_dir

path to Sources directory

Type

str

sources_names

list of names of available sources

Type

list[str]

targets_names

list of names of available targets

Type

list[str]

setup

loaded yaml configuration

Type

Dict

sample_rate

sets sample rate for optional resampling. Defaults to none which results in 44100.0

Type

Optional(Float)

load_mus_tracks()[source]

Iterates through the musdb folder structure and returns Track objects

download()[source]

Download the MUSDB Sample data

get_track_indices_by_names(names)[source]

Returns musdb track indices by track name

Can be used to filter the musdb tracks for a validation subset by trackname

Parameters == == == == == names : list[str], optional

select tracks by a given str or list of tracknames

Returns

return a list of Track Objects

Return type

list[int]

get_validation_track_indices(validation_track_names=None)[source]

Returns validation track indices by a given list of track names

Defaults to the builtin selection 8 validation tracks, defined in mus.yaml.

Parameters == == == == == validation_track_names : list[str], optional

validation track names by a given str or list of tracknames

Returns

return a list of validation track indices

Return type

list[int]

load_mus_tracks(subsets=None, split=None)[source]

Parses the musdb folder structure, returns list of Track objects

Parameters
  • subsets (list[str], optional) – select a _musdb_ subset train or test. Default None loads [train, test].

  • split (str) – for subsets=’train’, split=’train applies a train/validation split. if split=’valid’ the validation split of the training subset will be used

Returns

return a list of Track Objects

Return type

list[Track]

save_estimates(user_estimates, track, estimates_dir, write_stems=False)[source]

Writes user_estimates to disk while recreating the musdb file structure in that folder.

Parameters
  • user_estimates (Dict[np.array]) – the target estimates.

  • track (Track,) – musdb track object

  • estimates_dir (str,) – output folder name where to save the estimates.

class musdb.audio_classes.MultiTrack(path=None, name=None, artist=None, title=None, sources=None, targets=None, sample_rate=None, *args, **kwargs)[source]
property stems

[shape=(stems, num_samples, num_channels)]

Type

array_like

class musdb.audio_classes.Source(multitrack, name=None, path=None, stem_id=None, gain=1.0, *args, **kwargs)[source]

An audio Target which is a linear mixture of several sources

name

Name of this source

Type

str

stem_id

stem/substream ID is set here.

Type

int

is_wav

If stem is read from wav or mp4 stem

Type

boolean

path

Absolute path to audio file

Type

str

gain

Mixing weight for this source

Type

float

class musdb.audio_classes.Target(multitrack, sources, name=None)[source]

An audio Target which is a linear mixture of several sources

multitrack

Track object

Type

Track

sources

list of Source objects for this Target

Type

list[Source]

property audio

[shape=(num_samples, num_channels)]

mixes audio for targets on the fly

Type

array_like

class musdb.audio_classes.Track(path='None', is_wav=False, stem_id=None, subset=None, chunk_start=0, chunk_duration=None, sample_rate=None)[source]

Generic audio Track that can be wav or stem file

name

Track name

Type

str

path

Absolute path of mixture audio file

Type

str

stem_id

stem/substream ID

Type

int

is_wav

If stem is read from wav or mp4 stem

Type

boolean

subset

Track belongs to which subset.

Type

{‘train’, ‘test’}

targets

OrderedDict of mixted Targets for this track.

Type

OrderedDict

sources

Dict of Source objects for this track.

Type

Dict

chunk_start

sets offset when loading the audio, defaults to 0 (beginning).

Type

float

chunk_duration

sets duration for the audio, defaults to None (end).

Type

float

load_audio(path, stem_id, chunk_start=0, chunk_duration=None, sample_rate=None)[source]

array_like: [shape=(num_samples, num_channels)]

musdb.tools.musdb_convert(inargs=None)[source]

cli application to conver audio files to spectrogram images

Indices and tables

Citation