Utilities to convert from ground-truth

asmd.convert_from_file._sort_alignment(alignment, data)

Sort data in alignment (in-place)

asmd.convert_from_file._sort_lists(*lists)

Sort multiple lists in-place with reference to the first one

asmd.convert_from_file._sort_pedal(data)

Sort pedal for data (in-place)

asmd.convert_from_file.change_ext(input_fn, new_ext, no_dot=False, remove_player=False)

Return the input path input_fn with new_ext as extension and the part after the last ‘-’ removed. If no_dot is True, it will not add a dot before of the extension, otherwise it will add it if not present. remove_player can be used to remove the name of the player in the last part of the file name when: use this for the traditional_flute dataset; it will remove the last part after ‘_’.

asmd.convert_from_file.convert(exts, no_dot=True, remove_player=False)

This function is designed to be used as decorators for functions which converts from a filetype to our JSON format.

Example of usage:

>>> @convert(['.myext'], no_dot=True, remove_player=False)
... def function_which_converts(...):
...     pass

Parameters:

• ext (*) – the possible extensions of the ground-truths to be converted, e.g. [‘.mid’, ‘.midi’]. You can also use this parameter to remove exceeding parts at the end of the filename (see from_bach10_mat and from_bach10_f0 source code)
• no_dot (*) – if True, don’t add a dot before of the extension, if False, add it if not present; this is useful if you are using the extension to remove other parts in the file name (see ext).
• remove_player (*) – if True, remove the name of the player in the last part of the file name: use this for the traditional_flute dataset; it will remove the part after the last ‘_’.

asmd.convert_from_file.from_bach10_f0(nmat_fn, sources=range(0, 4))

Open a matlab mat file nmat_fn in the MIREX format (Bach10) for frame evaluation and convert it to our ground_truth representation. This fills: f0. sources is an iterable containing the indices of the sources to be considered, where the first source is 0. Returns a list of dictionary, one per source.

asmd.convert_from_file.from_bach10_mat(mat_fn, sources=range(0, 4))

Open a txt file txt_fn in the MIREX format (Bach10) and convert it to our ground_truth representation. This fills: precise_alignment, pitches. sources is an iterable containing the indices of the sources to be considered, where the first source is 0. Returns a list of dictionary, one per source.

asmd.convert_from_file.from_midi(midi_fn, alignment='precise_alignment', pitches=True, velocities=True, merge=True, beats=False)

Open a midi file midi_fn and convert it to our ground_truth representation. This fills velocities, pitches, beats, sustain, soft, sostenuto and alignment (default: precise_alignment). Returns a list containing a dictionary. alignment can also be None or False, in that case no alignment is filled. If merge is True, the returned list will contain a dictionary for each track. Beats are filled according to tempo changes.

This functions is decorated with 3 different sets of parameters:

• from_midi is the decorated version with remove_player=False
• from_midi_remove_player is the decorated version with remove_player=True
• from_midi_asap is the decorated version which accept extension ‘.score.mid’ which is used in the script to import scores from ASAP

N.B. To allow having some annotation for subgroups of a dataset, this function returns None when it cannot find the specified midi file; in this way, that file is not taken into account while merging the various annotations (e.g. asap group inside Maestro dataset)

asmd.convert_from_file.from_midi_asap(midi_fn, alignment='precise_alignment', pitches=True, velocities=True, merge=True, beats=False)

Open a midi file midi_fn and convert it to our ground_truth representation. This fills velocities, pitches, beats, sustain, soft, sostenuto and alignment (default: precise_alignment). Returns a list containing a dictionary. alignment can also be None or False, in that case no alignment is filled. If merge is True, the returned list will contain a dictionary for each track. Beats are filled according to tempo changes.

This functions is decorated with 3 different sets of parameters:

• from_midi is the decorated version with remove_player=False
• from_midi_remove_player is the decorated version with remove_player=True
• from_midi_asap is the decorated version which accept extension ‘.score.mid’ which is used in the script to import scores from ASAP

N.B. To allow having some annotation for subgroups of a dataset, this function returns None when it cannot find the specified midi file; in this way, that file is not taken into account while merging the various annotations (e.g. asap group inside Maestro dataset)

asmd.convert_from_file.from_midi_remove_player(midi_fn, alignment='precise_alignment', pitches=True, velocities=True, merge=True, beats=False)

Open a midi file midi_fn and convert it to our ground_truth representation. This fills velocities, pitches, beats, sustain, soft, sostenuto and alignment (default: precise_alignment). Returns a list containing a dictionary. alignment can also be None or False, in that case no alignment is filled. If merge is True, the returned list will contain a dictionary for each track. Beats are filled according to tempo changes.

This functions is decorated with 3 different sets of parameters:

• from_midi is the decorated version with remove_player=False
• from_midi_remove_player is the decorated version with remove_player=True
• from_midi_asap is the decorated version which accept extension ‘.score.mid’ which is used in the script to import scores from ASAP

N.B. To allow having some annotation for subgroups of a dataset, this function returns None when it cannot find the specified midi file; in this way, that file is not taken into account while merging the various annotations (e.g. asap group inside Maestro dataset)

asmd.convert_from_file.from_musicnet_csv(csv_fn, sr=44100.0)

Open a csv file csv_fn and convert it to our ground_truth representation. This fills: broad_alignment, score, pitches. This returns a list containing only one dict. sr is the samplerate of the audio files (MusicNet csv contains the sample number as onset and offsets of each note) and it shold be a float.

N.B. MusicNet contains wav files at 44100 Hz as samplerate. N.B. Lowest in pitch in musicnet is 21, so we assume that they count pitch starting with 0 as in midi.org standard. N.B. score times are provided with BPM 60 for all the scores

asmd.convert_from_file.from_phenicx_txt(txt_fn)

Open a txt file txt_fn in the PHENICX format and convert it to our ground_truth representation. This fills: broad_alignment.

asmd.convert_from_file.from_sonic_visualizer(gt_fn, alignment='precise_alignment')

Takes a filename of a sonic visualizer output file exported as ‘csv’ and fills the ‘alignment’ specified

asmd.convert_from_file.prototype_gt = {'broad_alignment': {'notes': [], 'offsets': [], 'onsets': [], 'pitches': [], 'velocities': []}, 'extra': [], 'f0': [], 'instrument': 255, 'misaligned': {'notes': [], 'offsets': [], 'onsets': [], 'pitches': [], 'velocities': []}, 'missing': [], 'precise_alignment': {'notes': [], 'offsets': [], 'onsets': [], 'pitches': [], 'velocities': []}, 'score': {'beats': [], 'notes': [], 'offsets': [], 'onsets': [], 'pitches': [], 'velocities': []}, 'soft': {'times': [], 'values': []}, 'sostenuto': {'times': [], 'values': []}, 'sustain': {'times': [], 'values': []}}

The dictionary prototype for containing the ground_truth. use:

>>> from copy import deepcopy
... from convert_from_file import prototype_gt
... prototype_gt = deepcopy(prototype_gt)

>>> prototype_gt
{
    "precise_alignment": {
        "onsets": [],
        "offsets": [],
        "pitches": [],
        "notes": [],
        "velocities": []
    },
    "misaligned": {
        "onsets": [],
        "offsets": [],
        "pitches": [],
        "notes": [],
        "velocities": []
    },
    "score": {
        "onsets": [],
        "offsets": [],
        "pitches": [],
        "notes": [],
        "velocities": [],
        "beats": []
    },
    "broad_alignment": {
        "onsets": [],
        "offsets": [],
        "pitches": [],
        "notes": [],
        "velocities": []
    },
    "f0": [],
    "soft": {
        "values": [],
        "times": []
    },
    "sostenuto": {
        "values": [],
        "times": []
    },
    "sustain": {
        "values": [],
        "times": []
    },
    "instrument": 255,
}

Note: pitches, velocities, sustain, sostenuto, soft, and (if available) instrument must be in range [0, 128)