ch_range=range(1, 9),
fs=48000)
# audio_data = get_audio_nochime('new_dataset/new_audio/AUDIO_RECORDING', ch_range=range(1, 9), fs=49000)
# calculate the time for load the audio files
t_io += t.msecs
# change the audio files into frequency domain
Y = stft(audio_data, time_dim=1).transpose((1, 0, 2))
print(audio_data.shape, type(audio_data))
Y_var = Variable(np.abs(Y).astype(np.float32), True)
# mask estimation
with Timer() as t:
N_masks, X_masks = model.calc_masks(Y_var)
N_masks.to_cpu()
X_masks.to_cpu()
t_net += t.msecs
with Timer() as t:
N_mask = np.median(N_masks.data, axis=1)
X_mask = np.median(X_masks.data, axis=1)
print("Y: ",
Y.shape,
"N_mask: ",
N_mask.shape,
"X_mask: ",
X_mask.shape,
end="\n")
Y_hat = gev_wrapper_on_masks(Y, N_mask, X_mask)
# Beamform loop
for cur_line in tqdm(flist):
with Timer() as t:
if scenario == 'simu':
audio_data = get_audio_data(cur_line)
context_samples = 0
elif scenario == 'real':
audio_data, context_samples = get_audio_data_with_context(
cur_line[0], cur_line[1], cur_line[2])
t_io += t.msecs
Y = stft(audio_data, time_dim=1).transpose((1, 0, 2))
Y_var = Variable(np.abs(Y).astype(np.float32))
if args.gpu >= 0:
Y_var.to_gpu(args.gpu)
with Timer() as t:
N_masks, X_masks = model.calc_masks(Y_var)
N_masks.to_cpu()
X_masks.to_cpu()
t_net += t.msecs
with Timer() as t:
N_mask = np.median(N_masks.data, axis=1)
X_mask = np.median(X_masks.data, axis=1)
Y_hat_dicts = bf_wrapper_on_masks(Y, N_mask, X_mask, beamformers=beamformers)
t_beamform += t.msecs
if scenario == 'simu':
wsj_name = cur_line.split('/')[-1].split('_')[1]
spk = cur_line.split('/')[-1].split('_')[0]
env = cur_line.split('/')[-1].split('_')[-1]
elif scenario == 'real':
# Beamform loop
for cur_line in tqdm(flist):
with Timer() as t:
if scenario == 'simu':
audio_data = get_audio_data(cur_line)
context_samples = 0
elif scenario == 'real':
audio_data, context_samples = get_audio_data_with_context(
cur_line[0], cur_line[1], cur_line[2])
t_io += t.msecs
Y = stft(audio_data, time_dim=1).transpose((1, 0, 2))
Y_var = Variable(np.abs(Y).astype(np.float32), True)
if args.gpu >= 0:
Y_var.to_gpu(args.gpu)
with Timer() as t:
N_masks, X_masks = model.calc_masks(Y_var)
N_masks.to_cpu()
X_masks.to_cpu()
t_net += t.msecs
with Timer() as t:
N_mask = np.median(N_masks.data, axis=1)
X_mask = np.median(X_masks.data, axis=1)
Y_hat = gev_wrapper_on_masks(Y, N_mask, X_mask)
t_beamform += t.msecs
if scenario == 'simu':
wsj_name = cur_line.split('/')[-1].split('_')[1]
spk = cur_line.split('/')[-1].split('_')[0]
env = cur_line.split('/')[-1].split('_')[-1]
elif scenario == 'real':
