def get_config(batch_size, data_path):
return configs.Config(
model=MusicVAE(lstm_models.BidirectionalLstmEncoder(),
lstm_models.CategoricalLstmDecoder()),
hparams=merge_hparams(
lstm_models.get_default_hparams(),
HParams(
batch_size=512,
max_seq_len=32, # 2 bars w/ 16 steps per bar
z_size=512,
enc_rnn_size=[2048],
dec_rnn_size=[2048, 2048, 2048],
free_bits=0,
max_beta=0.5,
beta_rate=0.99999,
sampling_schedule='inverse_sigmoid',
sampling_rate=1000,
)),
note_sequence_augmenter=data.NoteSequenceAugmenter(
transpose_range=(-5, 5)),
data_converter=data.OneHotMelodyConverter(
valid_programs=data.MEL_PROGRAMS,
skip_polyphony=False,
max_bars=100, # Truncate long melodies before slicing.
slice_bars=2,
steps_per_quarter=4),
train_examples_path=data_path,
eval_examples_path=data_path,
)
def magent_melody_extractor(midi_path, max_bars=100):
data_converter = music_vae_data.OneHotMelodyConverter(
valid_programs=music_vae_data.MEL_PROGRAMS,
skip_polyphony=False,
max_bars=max_bars, # Truncate long melodies before slicing.
slice_bars=16,
steps_per_quarter=4)
ns = convert_midi("", '', midi_path)
all_melodies = []
try:
note_sequences = sequences_lib.split_note_sequence_on_time_changes(ns)
for ns in note_sequences:
# filter notes --------------------------------------------------------
def filter(note):
if (data_converter._valid_programs is not None and note.program
not in data_converter._valid_programs):
return False
return data_converter._min_pitch <= note.pitch <= data_converter._max_pitch
notes = list(ns.notes)
del ns.notes[:]
ns.notes.extend([n for n in notes if filter(n)])
# quantize sequence ---------------------------------------------------
quantized_sequence = mm.quantize_note_sequence(
ns, data_converter._steps_per_quarter)
# extract melodies ----------------------------------------------------
melodies, _ = data_converter._event_extractor_fn(
quantized_sequence)
all_melodies.extend(melodies)
except:
pass
return all_melodies
def testIsTraining(self):
converter = data.OneHotMelodyConverter(
steps_per_quarter=1, slice_bars=2, max_tensors_per_notesequence=2)
self.is_training = True
self.assertEqual(2, len(converter.to_tensors(self.sequence)[0]))
converter.max_tensors_per_notesequence = None
self.assertEqual(5, len(converter.to_tensors(self.sequence)[0]))
def testIsTraining(self):
converter = data.OneHotMelodyConverter(
steps_per_quarter=1, slice_bars=2, max_tensors_per_notesequence=2)
converter.set_mode('train')
self.assertEqual(2, len(converter.to_tensors(self.sequence).inputs))
converter.max_tensors_per_notesequence = None
self.assertEqual(5, len(converter.to_tensors(self.sequence).inputs))
def testMaxOutputsPerNoteSequence(self):
converter = data.OneHotMelodyConverter(
steps_per_quarter=1, slice_bars=2, max_tensors_per_notesequence=2)
self.assertEqual(2, len(converter.to_tensors(self.sequence)[0]))
converter.max_tensors_per_notesequence = 3
self.assertEqual(3, len(converter.to_tensors(self.sequence)[0]))
converter.max_tensors_per_notesequence = 100
self.assertEqual(5, len(converter.to_tensors(self.sequence)[0]))
def testToNoteSequence(self):
converter = data.OneHotMelodyConverter(
steps_per_quarter=1, slice_bars=4, max_tensors_per_notesequence=1)
_, output_tensors = converter.to_tensors(
filter_instrument(self.sequence, 0))
sequences = converter.to_notesequences(output_tensors)
self.assertEqual(1, len(sequences))
expected_sequence = music_pb2.NoteSequence(ticks_per_quarter=220)
expected_sequence.tempos.add(qpm=120)
testing_lib.add_track_to_sequence(
expected_sequence, 0,
[(32, 80, 1.0, 2.0), (33, 80, 3.0, 5.5), (34, 80, 5.5, 6.5)])
self.assertProtoEquals(expected_sequence, sequences[0])
def testToNoteSequenceChordConditioned(self):
converter = data.OneHotMelodyConverter(
steps_per_quarter=1, slice_bars=4, max_tensors_per_notesequence=1,
chord_encoding=mm.MajorMinorChordOneHotEncoding())
tensors = converter.to_tensors(
filter_instrument(self.sequence, 0))
sequences = converter.to_notesequences(tensors.outputs, tensors.controls)
self.assertEqual(1, len(sequences))
expected_sequence = music_pb2.NoteSequence(ticks_per_quarter=220)
expected_sequence.tempos.add(qpm=120)
testing_lib.add_track_to_sequence(
expected_sequence, 0,
[(32, 80, 1.0, 2.0), (33, 80, 3.0, 5.5), (34, 80, 5.5, 6.5)])
testing_lib.add_chords_to_sequence(
expected_sequence, [('N.C.', 0), ('F', 1), ('C', 4)])
self.assertProtoEquals(expected_sequence, sequences[0])
batch_size=512,
max_seq_len=32, # 2 bars w/ 16 steps per bar
z_size=256,
enc_rnn_size=[512],
dec_rnn_size=[256, 256],
free_bits=0,
max_beta=0.2,
beta_rate=0.99999,
sampling_schedule='inverse_sigmoid',
sampling_rate=1000,
)),
note_sequence_augmenter=data.NoteSequenceAugmenter(transpose_range=(-5,
5)),
data_converter=data.OneHotMelodyConverter(
valid_programs=data.MEL_PROGRAMS,
skip_polyphony=False,
max_bars=100, # Truncate long melodies before slicing.
slice_bars=2,
steps_per_quarter=4),
train_examples_path=None,
eval_examples_path=None,
)
CONFIG_MAP['cat-mel_2bar_big'] = Config(
model=MusicVAE(lstm_models.BidirectionalLstmEncoder(),
lstm_models.CategoricalLstmDecoder()),
hparams=merge_hparams(
lstm_models.get_default_hparams(),
HParams(
batch_size=512,
max_seq_len=32, # 2 bars w/ 16 steps per bar
z_size=512,
config_map['cat-mel_2bar_small'] = Config(
model=MusicVAE(lstm_models.BidirectionalLstmEncoder(),
lstm_models.CategoricalLstmDecoder()),
hparams=merge_hparams(
lstm_models.get_default_hparams(),
HParams(
batch_size=512,
max_seq_len=32, # 2 bars w/ 16 steps per bar
z_size=256,
enc_rnn_size=[512],
dec_rnn_size=[256, 256],
)),
note_sequence_augmenter=None,
note_sequence_converter=data.OneHotMelodyConverter(
valid_programs=data.MEL_PROGRAMS,
skip_polyphony=True,
max_bars=100, # Truncate long melodies before slicing.
slice_bars=2,
steps_per_quarter=4),
train_examples_path=None,
eval_examples_path=None,
)
config_map['cat-mel_2bar_big'] = Config(
model=MusicVAE(lstm_models.BidirectionalLstmEncoder(),
lstm_models.CategoricalLstmDecoder()),
hparams=merge_hparams(
lstm_models.get_default_hparams(),
HParams(
batch_size=512,
max_seq_len=32, # 2 bars w/ 16 steps per bar
z_size=512,