The gan-voice-conversion from njellinas

CycleGAN-VC2

Output filter dimension in 2-1-2D CNN layer

Hi,
You have used the number of filters as 1 instead of 35 as specified in the original paper. Is there a reason for that or do you believe the architecture specified in the paper is off?

GAN-Voice-Conversion/models/modules.py

Line 299 in 1cfa34d

    
           conv_out = conv2d_layer(inputs=u2, filters=1, kernel_size=[5, 15], strides=[1, 1], activation=None,

'dim' parameter not using

GAN-Voice-Conversion/models/modules.py

Line 255 in 1cfa34d

    
           def Generator212(inputs, dim=24, batch_size=1, reuse=False, scope_name='generator_212'):

Changing num_mcep value always results in an error

Traceback (most recent call last):
  File "train_cyclegan_vc2.py", line 39, in <module>
    model = CycleGAN2(num_features=num_mcep, batch_size=mini_batch_size, log_dir=log_dir)
  File "/media/dmitriy/Main2/NeuralNetworks/CycleGAN-VC2-2/GAN-Voice-Conversion/models/cyclegan_vc2.py", line 21, in __init__
    self.build_model()
  File "/media/dmitriy/Main2/NeuralNetworks/CycleGAN-VC2-2/GAN-Voice-Conversion/models/cyclegan_vc2.py", line 81, in build_model
    self.cycle_loss = l1_loss(y=self.input_A_real, y_hat=self.cycle_A) + l1_loss(y=self.input_B_real,
  File "/media/dmitriy/Main2/NeuralNetworks/CycleGAN-VC2-2/GAN-Voice-Conversion/models/utils.py", line 5, in l1_loss
    return tf.reduce_mean(tf.abs(y - y_hat))
  File "/media/dmitriy/Main2/NeuralNetworks/Anaconda3/envs/cycleganvc/lib/python3.7/site-packages/tensorflow/python/ops/math_ops.py", line 812, in binary_op_wrapper
    return func(x, y, name=name)
  File "/media/dmitriy/Main2/NeuralNetworks/Anaconda3/envs/cycleganvc/lib/python3.7/site-packages/tensorflow/python/ops/gen_math_ops.py", line 9536, in sub
    "Sub", x=x, y=y, name=name)
  File "/media/dmitriy/Main2/NeuralNetworks/Anaconda3/envs/cycleganvc/lib/python3.7/site-packages/tensorflow/python/framework/op_def_library.py", line 788, in _apply_op_helper
    op_def=op_def)
  File "/media/dmitriy/Main2/NeuralNetworks/Anaconda3/envs/cycleganvc/lib/python3.7/site-packages/tensorflow/python/util/deprecation.py", line 507, in new_func
    return func(*args, **kwargs)
  File "/media/dmitriy/Main2/NeuralNetworks/Anaconda3/envs/cycleganvc/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 3300, in create_op
    op_def=op_def)
  File "/media/dmitriy/Main2/NeuralNetworks/Anaconda3/envs/cycleganvc/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 1823, in __init__
    control_input_ops)
  File "/media/dmitriy/Main2/NeuralNetworks/Anaconda3/envs/cycleganvc/lib/python3.7/site-packages/tensorflow/python/framework/ops.py", line 1662, in _create_c_op
    raise ValueError(str(e))
ValueError: Dimensions must be equal, but are 70 and 36 for 'sub' (op: 'Sub') with input shapes: [?,70,?], [1,36,?].

I've noticed that on the leimao's implementation the num_mcep value is 24 but on yours it's 36. Can you please explain how you changed it? It seems to always give me an error when i try.

njellinas / gan-voice-conversion Goto Github PK

gan-voice-conversion's People

Contributors

Stargazers

Watchers

Forkers

gan-voice-conversion's Issues

CycleGAN-VC2

Output filter dimension in 2-1-2D CNN layer

'dim' parameter not using

Changing num_mcep value always results in an error

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent