def parse_args(argv):
parser = common.common_arguments({
"samplerate": 44100, "context_width": 5*HOP_LENGTH, "annotations_per_window": 10, "hop_size": 1, "frame_width": HOP_LENGTH,
"note_range": 72, "min_note": 24, "batch_size": 32,
"evaluate_every": 5000,
"evaluate_small_every": 1000,
})
# Model specific arguments
parser.add_argument("--spectrogram", default="cqt", type=str, help="Postprocessing layer")
parser.add_argument("--first_pool_type", default=None, type=str, help="First pooling type")
parser.add_argument("--first_pool_size", default=[1, 5], nargs="+", type=str, help="Input pooling size")
parser.add_argument("--first_pool_stride", default=[1, 5], nargs="+", type=str, help="Input pooling stride")
parser.add_argument("--capacity_multiplier", default=8, type=int)
parser.add_argument("--architecture", default="full_1layer", type=str)
parser.add_argument("--conv_ctx", default=0, type=int)
parser.add_argument("--batchnorm", default=0, type=int)
parser.add_argument("--dropout", default=0.0, type=float)
parser.add_argument("--last_layer", default="conv", type=str)
parser.add_argument("--last_conv_ctx", default=0, type=int)
parser.add_argument("--harmonic_stacking", default=1, type=int)
parser.add_argument("--activation", default="relu", type=str)
args = parser.parse_args(argv)
common.name(args, "voicing")
return args
def parse_args(argv):
parser = common.common_arguments({"context_width": 978})
parser.add_argument("--capacity_multiplier",
default=16,
type=int,
help="Capacity multiplier of the model")
parser.add_argument(
"--multiresolution_convolution",
default=0,
type=int,
help="Number of different resolution of the first convolution layer")
parser.add_argument("--variable_stride",
action='store_true',
default=False,
help="Variable stride")
parser.add_argument("--first_layer_capacity",
default=1,
type=int,
help="Capacity multiplier")
args = parser.parse_args(argv)
common.name(args, "crepe")
return args
def parse_args(argv):
hop_length = 512
parser = common.common_arguments({
"samplerate": 44100,
"context_width": 10 * hop_length,
"annotations_per_window": 20,
"hop_size": 1,
"frame_width": hop_length,
"note_range": 72,
"min_note": 24,
"evaluate_every": 5000,
"evaluate_small_every": 1000,
})
# Model specific arguments
parser.add_argument("--spectrogram",
default="cqt",
type=str,
help="Postprocessing layer")
parser.add_argument("--capacity_multiplier",
default=8,
type=int,
help="Capacity")
parser.add_argument("--voicing_capacity_multiplier",
default=8,
type=int,
help="Capacity")
parser.add_argument("--undertone_stacking",
default=1,
type=int,
help="spectrogram stacking")
parser.add_argument("--overtone_stacking",
default=5,
type=int,
help="spectrogram stacking")
parser.add_argument("--voicing",
action='store_true',
help="Add voicing model.")
parser.add_argument("--conv_ctx", default=0, type=int)
parser.add_argument("--last_conv_ctx", default=0, type=int)
parser.add_argument("--voicing_conv_ctx", default=0, type=int)
parser.add_argument("--voicing_last_conv_ctx", default=0, type=int)
parser.add_argument("--batchnorm", default=0, type=int)
parser.add_argument("--dropout", default=0.3, type=float)
parser.add_argument("--activation", default="relu", type=str)
args = parser.parse_args(argv)
common.name(args, "cqt_voicing_residual_batchnorm")
return args
def parse_args(argv):
parser = common.common_arguments({
"samplerate": 22050, "context_width": 0, "annotations_per_window": 50, "hop_size": 1
, "frame_width": HOP_LENGTH,
"note_range": 72, "min_note": 24,
"evaluate_every": 5000,
"evaluate_small_every": 1000,
})
# Model specific arguments
parser.add_argument("--spectrogram", default="cqt", type=str, help="Postprocessing layer")
args = parser.parse_args(argv)
common.name(args, "bittner")
return args
def parse_args(argv):
parser = common.common_arguments({"context_width": 0})
# Model specific arguments
parser.add_argument("--initial_filter_width", default=32, type=int, help="First conv layer filter width")
parser.add_argument("--initial_filter_padding", default="same", type=str, help="First conv layer padding")
parser.add_argument("--filter_width", default=3, type=int, help="Dilation stack filter width (2 or 3)")
parser.add_argument("--use_biases", action='store_true', default=False, help="Use biases in the convolutions")
parser.add_argument("--skip_channels", default=64, type=int, help="Skip channels")
parser.add_argument("--residual_channels", default=32, type=int, help="Residual channels")
parser.add_argument("--stack_number", default=1, type=int, help="Number of dilated stacks")
parser.add_argument("--max_dilation", default=512, type=int, help="Maximum dilation rate")
parser.add_argument("--dilation_layer_dropout", default=0.0, type=float, help="Dropout in dilation layer")
parser.add_argument("--skip_layer_dropout", default=0.0, type=float, help="Dropout in skip connections")
parser.add_argument("--postprocessing", default="avgpool_p93_s93_Psame->conv_f256_k16_s8_Psame_arelu->conv_f256_k16_s8_Psame_arelu", type=str, help="Postprocessing layer")
args = parser.parse_args(argv)
common.name(args, "wavenet")
return args
def parse_args(argv):
hop_length = 512
parser = common.common_arguments({
"samplerate": 44100,
"context_width": 1 * hop_length,
"annotations_per_window": 1,
"hop_size": 1,
"frame_width": hop_length,
"note_range": 72,
"min_note": 24,
"evaluate_every": 5000,
"evaluate_small_every": 1000,
"batch_size": 1,
"annotation_smoothing": 0.18,
"datasets": ["mdb_mel4"]
})
# Model specific arguments
parser.add_argument("--spectrogram",
default="cqt_fs",
type=str,
help="Postprocessing layer")
parser.add_argument("--architecture",
default="bittnerlike",
type=str,
help="Postprocessing layer")
parser.add_argument("--capacity_multiplier",
default=8,
type=int,
help="Capacity")
parser.add_argument("--stacks", default=10, type=int, help="Stacks")
parser.add_argument("--conv_range",
default=3,
type=int,
help="Stack kernel width")
parser.add_argument("--harmonic_stacking",
default=1,
type=int,
help="harmonic stacking undertones and overtones")
parser.add_argument("--voicing_capacity_multiplier",
default=8,
type=int,
help="Capacity")
parser.add_argument("--undertone_stacking",
default=5,
type=int,
help="spectrogram stacking")
parser.add_argument("--overtone_stacking",
default=10,
type=int,
help="spectrogram stacking")
parser.add_argument("--voicing",
action='store_true',
help="Add voicing model.")
parser.add_argument("--conv_ctx", default=0, type=int)
parser.add_argument("--last_conv_ctx", default=0, type=int)
parser.add_argument("--voicing_conv_ctx", default=0, type=int)
parser.add_argument("--voicing_last_conv_ctx", default=0, type=int)
parser.add_argument("--batchnorm", default=0, type=int)
parser.add_argument("--dropout", default=0.3, type=float)
parser.add_argument("--activation", default="relu", type=str)
args = parser.parse_args(argv)
common.name(args, "cqtmf0")
return args