def play_rnn_OLA(model, path, n, win, format='rov', args=None, translation=True, lpf=True):
net = model(*args)
net.load_state_dict(torch.load(path))
x = net.generate(1, n).detach().squeeze().numpy()
x = iflatten_complex_data_with(x, 31)
x = pad_data_zeros(x, get_single_side_frequency().shape[0])
x = ifft_data(x)
x = iwindow(x, win + 0.001, 0.9)
x = istandardize_data(x)
x = overlap_and_add_data(x)
if lpf:
x = low_pass_filter(x)
write_one_file('example', x, format=format)
cmd = CMD + PAR1 + 'example' + PAR2
if translation:
cmd += TRANSLATION
os.system(cmd)
return x
def example(self):
# Generate an example with numpy array data type
x = self.generate(1).detach().cpu()
x = np.array(x)
x = iflatten_complex_data_with(x, 16)
x = pad_data_zeros(x, get_single_side_frequency().shape[0])
x = ifft_data(x)
# x = iwindow(x, WIN, 0.8)
return x[0]
def sqrt_spectra(examples):
n = examples[0].shape[0]
freq = get_single_side_frequency(n)
spectras = []
fe = fft_data(examples)
for e in fe:
spectras.append((freq, np.absolute(e)))
return spectras
def log_spectra(examples):
n = examples[0].shape[0]
freq = get_single_side_frequency(n)
spectras = []
fe = fft_data(examples)
for e in fe:
m = np.amax(np.absolute(e))
spectras.append((freq, 20 * np.log10(np.absolute(e) / m)))
return spectras
def example(self, r):
# Generate an example with numpy array data type
global dim
x = self.generate(1, r).detach().cpu()
x = x.squeeze().unsqueeze(0)
x = np.array(x)
x = iflatten_complex_data_with(x, 31)
x = pad_data_zeros(x, get_single_side_frequency().shape[0])
x = ifft_data(x)
# x = iwindow(x, WIN, 0.8)
return x[0]
def lpf_dimension_reduction(data, frequency, trim_length=None):
if trim_length is None:
trim_length = config.TRIM_LENGTH
bins = data[0].shape[0]
freq = get_single_side_frequency(trim_length)
idx = freq <= frequency
dim = np.sum(idx)
data_reducted = []
for d in data:
data_reducted.append(d[idx])
return data_reducted, dim
def _draw(i):
global AX1
global AX2
global AX3
global AX4
global FIGURE
global QUEUE
global DATA
if not QUEUE.empty():
DATA = QUEUE.get()
if DATA == 'close':
plt.close(FIGURE)
return
example_length = DATA['example'].shape[0]
freq = get_single_side_frequency(example_length)
# DATA['loss_title'] = data['loss_title']
# DATA['losses'].append(data['loss']),
# DATA['loss_labels'] = data['loss_labels']
# DATA['score_title'] = data['score_title']
# DATA['scores'].append(data['score'])
# DATA['score_labels'] = data['score_labels']
# DATA['interval'] = data['interval']
n = len(DATA['losses'])
interval = DATA['interval']
x = np.arange(interval, interval * n + 1, interval)
loss_Y = np.array(DATA['losses']).T
score_Y = np.array(DATA['scores']).T
AX1.clear()
AX2.clear()
AX3.clear()
AX4.clear()
AX1.set_title(DATA['loss_title'])
AX1.set_xlabel('Training Steps')
AX1.set_ylabel('Model Loss')
for i in range(loss_Y.shape[0]):
AX1.plot(x, loss_Y[i].flatten(), label=DATA['loss_labels'][i])
AX1.legend()
AX2.set_title(DATA['score_title'])
AX2.set_xlabel('Training Steps')
AX2.set_ylabel('Score')
for i in range(score_Y.shape[0]):
AX2.plot(x, score_Y[i].flatten(), label=DATA['score_labels'][i])
AX2.legend()
AX3.set_title('Spectra')
AX3.set_xlabel('Frequency')
AX3.set_ylabel('Power (dB)')
example = DATA['example']
spectra = np.absolute(fft_data([DATA['example']])[0]).real
m = np.amax(spectra)
spectra = 20 * np.log10(spectra / m)
AX3.plot(freq, spectra)
AX4.set_title('Example')
AX4.set_xlabel('Time')
AX4.set_ylabel('Axis')
AX4.plot(example.real)
return