cell_dropout = .925
#noise_scale = 8.
prenet_units = 128
n_filts = 128
n_stacks = 3
enc_units = 128
dec_units = 512
emb_dim = 15
truncation_len = seq_len
cell_dropout_scale = cell_dropout
epsilon = 1E-8
forward_init = "truncated_normal"
rnn_init = "truncated_normal"
basedir = "/Tmp/kastner/lj_speech/LJSpeech-1.0/"
ljspeech = rsync_fetch(fetch_ljspeech, "leto01")
# THESE ARE CANNOT BE PAIRED (SOME MISSING), ITERATOR PAIRS THEM UP BY NAME
wavfiles = ljspeech["wavfiles"]
jsonfiles = ljspeech["jsonfiles"]
model_path = sys.argv[1]
seed = int(abs(hash(model_path))) % (2**32 - 1)
# THESE HAVE TO BE THE SAME TO ENSURE SPLIT IS CORRECT
train_random_state = np.random.RandomState(seed)
valid_random_state = np.random.RandomState(seed)
train_itr = wavfile_caching_mel_tbptt_iterator(wavfiles,
jsonfiles,
batch_size,
default=10,
type=int)
parser.add_argument('--output_mixtures',
dest='output_mixtures',
default=20,
type=int)
parser.add_argument('--lstm_layers', dest='lstm_layers', default=3, type=int)
parser.add_argument('--cell_dropout',
dest='cell_dropout',
default=.9,
type=float)
parser.add_argument('--units_per_layer', dest='units', default=400, type=int)
parser.add_argument('--restore', dest='restore', default=None, type=str)
args = parser.parse_args()
iamondb = rsync_fetch(fetch_iamondb, "leto01")
trace_data = iamondb["data"]
char_data = iamondb["target"]
batch_size = args.batch_size
truncation_len = args.seq_len
cell_dropout_scale = args.cell_dropout
vocabulary_size = len(iamondb["vocabulary"])
itr_random_state = np.random.RandomState(2177)
itr = tbptt_list_iterator(trace_data, [char_data],
batch_size,
truncation_len,
other_one_hot_size=[vocabulary_size],
random_state=itr_random_state)
epsilon = 1E-8
h_dim = args.units
def main():
iamondb = rsync_fetch(fetch_iamondb, "leto01")
translation = iamondb["vocabulary"]
#with open(os.path.join('data', 'translation.pkl'), 'rb') as file:
# translation = pickle.load(file)
rev_translation = {v: k for k, v in translation.items()}
charset = [rev_translation[i] for i in range(len(rev_translation))]
# just for display purposes - replace <NULL> with ''
charset[translation["<NULL>"]] = ""
assert translation["<NULL>"] == 0
config = tf.ConfigProto(device_count={'GPU': 0})
with tf.Session(config=config) as sess:
saver = tf.train.import_meta_graph(direct_model + '.meta')
saver.restore(sess, direct_model)
if args.text is not None:
args_text = args.text
else:
raise ValueError("Must pass --text argument")
phi_data, window_data, kappa_data, stroke_data, coords = sample_text(
sess, args_text, translation)
strokes = np.array(stroke_data)
epsilon = 1e-8
strokes[:, :2] = np.cumsum(strokes[:, :2], axis=0)
minx, maxx = np.min(strokes[:, 0]), np.max(strokes[:, 0])
miny, maxy = np.min(strokes[:, 1]), np.max(strokes[:, 1])
if args.info:
delta = abs(maxx - minx) / 400.
x = np.arange(minx, maxx, delta)
y = np.arange(miny, maxy, delta)
x_grid, y_grid = np.meshgrid(x, y)
z_grid = np.zeros_like(x_grid)
for i in range(strokes.shape[0]):
# what
cov = np.array([[strokes[i, 2], 0.], [0., strokes[i, 3]]])
gauss = mlab.bivariate_normal(x_grid,
y_grid,
mux=strokes[i, 0],
muy=strokes[i, 1],
sigmax=cov[0, 0],
sigmay=cov[1, 1],
sigmaxy=strokes[i, 4] *
cov[0, 0] * cov[1, 1])
# needs to be rho * sigmax * sigmay
z_grid += gauss * np.power(strokes[i, 2] + strokes[i, 3],
0.4) / (np.max(gauss) + epsilon)
for f in os.listdir("."):
if "plot" in f and f.endswith(".png"):
print("Removing old plot {}".format(f))
os.remove(f)
t = int(time.time())
new = "plot_{}".format(hash(t) % 10**5) + "_{}.png"
plt.figure()
plt.imshow(z_grid, interpolation="bilinear", cmap=cm.jet)
plt.axes().get_xaxis().set_visible(False)
plt.axes().get_yaxis().set_visible(False)
plt.axis("off")
new_d = new.format("density")
plt.title("Density")
print("Saving to {}".format(new_d))
plt.savefig(new_d)
plt.close()
plt.figure()
for stroke in split_strokes(cumsum(np.array(coords))):
if args.color is not None:
plt.plot(stroke[:, 0], -stroke[:, 1], color=args.color)
else:
plt.plot(stroke[:, 0], -stroke[:, 1])
plt.title(args.text)
plt.axes().set_aspect('equal')
plt.axes().get_xaxis().set_visible(False)
plt.axes().get_yaxis().set_visible(False)
plt.axis("off")
new_h = new.format("handwriting")
print("Saving to {}".format(new_h))
plt.savefig(new_h)
plt.close()
plt.figure()
phi_img = np.vstack(phi_data).T[::-1, :]
plt.imshow(phi_img,
interpolation='nearest',
aspect='auto',
cmap=cm.jet)
plt.yticks(np.arange(0, len(args_text) + 1))
plt.axes().set_yticklabels(list(' ' + args_text[::-1]),
rotation='vertical',
fontsize=8)
plt.grid(False)
plt.title('Phi')
new_p = new.format("phi")
print("Saving to {}".format(new_p))
plt.savefig(new_p)
plt.close()
plt.figure()
window_img = np.vstack(window_data).T
plt.imshow(window_img,
interpolation='nearest',
aspect='auto',
cmap=cm.jet)
plt.yticks(np.arange(0, len(charset)))
plt.axes().set_yticklabels(list(charset),
rotation='vertical',
fontsize=8)
plt.grid(False)
plt.title('Window')
new_w = new.format("window")
print("Saving to {}".format(new_w))
plt.savefig(new_w)
plt.close()
else:
fig, ax = plt.subplots(1, 1)
for stroke in split_strokes(cumsum(np.array(coords))):
if args.color is not None:
plt.plot(stroke[:, 0], -stroke[:, 1], color=args.color)
else:
plt.plot(stroke[:, 0], -stroke[:, 1])
ax.set_title(args.text)
ax.set_aspect('equal')
for f in os.listdir("."):
if "gen_plot" in f and f.endswith(".png"):
print("Removing old plot {}".format(f))
os.remove(f)
t = int(time.time())
new = "gen_plot_{}.png".format(hash(t) % 10**5)
print("Saving to {}".format(new))
plt.savefig(new)