os.path.join("./out", opt.opath),
"ae_n{}_b{}_{}".format(opt.n_fft, opt.n_mels, middle_size))
print('final output-path: {}'.format(statepath))
os.makedirs(statepath, exist_ok=True)
# log parameters
log_file = open(os.path.join(statepath, "params.txt"), "w")
log_file.write(str(opt))
log_file.close()
dset = SoundfileDataset(ipath=ipath, out_type="ae", normalize=True)
if DEBUG:
print('warning, debugging turnned on!')
dset.data = dset.data[:100]
tset, vset = dset.get_split(sampler=False, split_size=0.2)
TLoader = DataLoader(tset,
batch_size=batch_size,
shuffle=True,
drop_last=True,
num_workers=num_workers)
VLoader = DataLoader(vset,
batch_size=batch_size,
shuffle=False,
drop_last=True,
num_workers=num_workers)
vae = AutoEncoder(n_mels, encode=encode_size, middle=middle_size)
batch_size=opt.batch_size,
shuffle=True,
num_workers=int(opt.workers))
Iset = DatasetCust(opt.dataroot,
transform=transforms.Compose([
transforms.ToPILImage(),
transforms.Resize((opt.image_size, opt.image_size)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5),
(0.5, 0.5, 0.5)),
]))
assert Iset
if opt.debug:
Mset.data = Mset.data[:100]
assert len(Iset) > len(Mset)
Iset.data = Iset.data[:len(Mset)]
assert len(Iset) == len(Mset)
Iloader = torch.utils.data.DataLoader(Iset,
batch_size=opt.batch_size,
shuffle=True,
num_workers=int(opt.workers))
# Optimizers
optimizer_D = torch.optim.Adam(discriminator.parameters(),
lr=opt.lrD,
betas=(opt.b1, opt.b2))
optimizer_G = torch.optim.Adam(generator.parameters(),
lr=opt.lrG,
datapath = "./mels_set_f{}_h{}_b{}".format(n_fft, hop_length, n_mels)
statepath = "./lstm_f{}_h{}_b{}_no_max".format(n_fft, hop_length, n_mels)
#statepath = "conv_small_b128"
device = "cuda"
filt_genre = None
#filt_genre = ['Experimental', 'Instrumental', 'International', 'Pop']
dset = SoundfileDataset("./all_metadata.p",
ipath=datapath,
out_type="mel",
normalize=NORMALIZE,
n_time_steps=n_time_steps,
filter_list=filt_genre)
if DEBUG:
dset.data = dset.data[:2000]
tset, vset = dset.get_split(sampler=False)
TLoader = DataLoader(tset,
batch_size=batch_size,
shuffle=True,
drop_last=True,
num_workers=num_workers)
VLoader = DataLoader(vset,
batch_size=batch_size,
shuffle=False,
drop_last=True,
num_workers=num_workers)
model = LSTM(n_mels, batch_size, num_layers=n_layers)