jjery2243542 / voice_conversion Goto Github PK

When I try to run the test script, I get the following error. I do not have CUDA installed and I am using Windows 10. Is there something I can do to remove this ?
Thanks

may I ask where is the convert.py?

Hi @jjery2243542

Thank you for your code. When I read the code, I find there is something wrong:

For master branch:

No 'train' mode in function 'solver.Solver.train', but in main.py, I see the code 'solver.train(args.output_model_path, args.flag, mode='train')'.

Please have a check if you have time.

Hi, Thank you for the code. I am not able to get the length.txt file you use to create the json files. Also the implementation you've shared doesn't seem to have a training routine for the stage 1 vae can you share the training details for stage one followed by stage two. It would be massively helpful if you can include step wise training procedure in your read me.

Hello,

I was able to run make_dataset_vctk.py and get a h5py output. But the mcep normalize fails:
python mcep_normalize.py VCTK_h5py VCTK_h5py_norm
processing speaker_id=225
Traceback (most recent call last):
File "mcep_normalize.py", line 13, in
utt_f0 = [f_in[f'train/{speaker}/{utt_id}/log_f0'][()] for utt_id in f_in[f'train/{speaker}'].keys()]
File "mcep_normalize.py", line 13, in
utt_f0 = [f_in[f'train/{speaker}/{utt_id}/log_f0'][()] for utt_id in f_in[f'train/{speaker}'].keys()]
File "h5py/_objects.pyx", line 54, in h5py._objects.with_phil.wrapper
File "h5py/_objects.pyx", line 55, in h5py._objects.with_phil.wrapper
File "/home/tfs/venv_voice_conversion_jjerry/lib/python3.6/site-packages/h5py/_hl/group.py", line 177, in getitem
oid = h5o.open(self.id, self._e(name), lapl=self._lapl)
File "h5py/_objects.pyx", line 54, in h5py._objects.with_phil.wrapper
File "h5py/_objects.pyx", line 55, in h5py._objects.with_phil.wrapper
File "h5py/h5o.pyx", line 190, in h5py.h5o.open
KeyError: "Unable to open object (object 'log_f0' doesn't exist)"

Apparently, you are saving mel and lin in the make_dataset*.py modules. but you are reading log_f0, mc_mean, mc_std, f0_mean, f0_std in your mcep_normalize as well as your convert.py codes.

Is your documentation missing a step that arrives at log_f0, mc_mean, mc_std, f0_mean, f0_std ?

Please clarify

Thanks,
Buvana

Dear Developer,
Where is the .h5 file whose path is defined inside the code ? Do we need to compute it some way or is it present with the code?

Hi. I think your code has a CUDA dependency. However, I do not have CUDA or any GPU available. When I load my own saved model, I get the error "Torch not compiled with CUDA enabled" . Can you please modify the inference.py code so that I could test my model?
Thanks for your help.

The main.py file has been running for the past 16 hours and I also specified the output model path but there is not any model that is saved in that path ?
Am I missing something ? How do I save the model?
I am getting this window

Please provide the file which is referred in -index paththe path of sampled segment indexes (.json)..

or help me solve this issue when I get this error while running make_single_samples.py?

I was trying to run the pre-trained model but there is a file whose description is unknown and that is vctk.json. Did you already provide this file or is it missing ?

Hi @jjery2243542
According to new documentation, there is no preprocess/vctk.config file present and neither is the preprocess.sh file?
Am I missing something ?

Hello
May I ask how to convert raw .wav files into input format and why you don't use classify.py?

Hello,
I was trying to run the main.py file to retrain the model as you guided but in the middle of the process it threw an error "too many values to unpack (expected 2)" at line 279 in solver.py which is (c_i, c_j), (x_i_t, x_i_tk, x_i_prime, x_j) = self.permute_data(data).
Then I recognized that permute() function does not output many elements like this.

Did I miss something?
Can anybody help me out of it?

When I run the make_dataset_vctk.py script, I get the following error. I am using the following file as the speaker-info.txt. The error message is also attached to this.
Am I missing something?
speaker-info.txt

I am trying to run the make_single_samples.py and not sure what to pass in the

1. n_samples
2. segment length
3. training sampled index path (.json)

Can you please help ?
print('usage: python3 make_single_samples.py [h5py path] [training sampled index path (.json)] '
'[n_samples] [segment length] [used speaker file path]')

Hi @jjery2243542

I found in function solver.cal_loss:

'logits' dimention is: [batch, n_spkers * enc_len]
'y_true' dimention is: [batch, n_spkers]

Why this mimatch can work well?

Thank you

Hi dears friends

try this repository, it worked for me : https://github.com/arshd91/multitarget-voice-conversion-vctk

Could you provide this file, it is important to know the id of target speakers.