def __init__(self, roll=False, return_cqt=False, output=None, drop='last'):
super(PyTorch, self).__init__()
self.cqt = Spectrogram.CQT2010v2(
sr=44100,
hop_length=64,
n_bins=84*10,
bins_per_octave=12*10,
norm=1,
window='hann',
pad_mode='constant',
trainable=False,
)
self.roll = roll
self.drop = drop
self.return_cqt = return_cqt
self.output = output
self.bn0 = nn.BatchNorm1d(840)
self.conv1 = nn.Conv1d(840, 1024, 1)
self.bn1 = nn.BatchNorm1d(1024)
self.conv2 = nn.Conv1d(1024, 512, 1)
self.bn2 = nn.BatchNorm1d(512)
self.conv3 = nn.Conv1d(512, 256, 1)
self.bn3 = nn.BatchNorm1d(256)
self.conv4 = nn.Conv1d(256, 48, 1)
self.bn4 = nn.BatchNorm1d(48)
self.LSTM = nn.LSTM(48, 512, batch_first=True, num_layers=2, dropout=0.25)
self.conv5 = nn.Conv1d(512, 48, 1)
self.dropout_1 = nn.Dropout(p=0.25)
self.dropout_2 = nn.Dropout(p=0.25)
self.dropout_3 = nn.Dropout(p=0.25)
self.dropout_4 = nn.Dropout(p=0.25)
device = "cpu"
print("using CPU")
elif args.device == "GPU":
device = "cuda:0"
print("using GPU")
elif args.device == "librosa":
print("using librosa")
y_list = np.load(Path(__file__).parent / './y_list.npy')
if args.device in ["CPU", "GPU"]:
y_torch = torch.tensor(y_list, device=device).float()
spec_layer = Spectrogram.CQT2010v2(sr=44100,
n_bins=84,
bins_per_octave=24,
fmin=55,
device=device)
timing = []
for e in range(20):
t_start = time.time()
spec = spec_layer(y_torch[:1000])
spec = spec_layer(y_torch[1000:])
time_used = time.time() - t_start
# print(time_used)
timing.append(time_used)
print("mean = ", np.mean(timing))
print("std = ", np.std(timing))
data = pd.DataFrame(timing, columns=['t_avg'])
