deepakbaby/isegan

Improved Speech Enhancement GANs

isegan's Issues

pre-emphasis/de-emphasis

Hi @deepakbaby :

There is trainable pre-emphasis layer, but as far as I see, trainable de-emphasis is missing, isn't it?

prepare_data problem

When I'm running python prepare_data.py, it appeared these messages.

Traceback (most recent call last):
File "prepare_data.py", line 98, in
hdf5storage.savemat(destinationfilenameclean, matcontent)
File "/home/snow/venv/lib/python3.6/site-packages/hdf5storage/init.py", line 1681, in savemat
options=options)
File "/home/snow/venv/lib/python3.6/site-packages/hdf5storage/init.py", line 1319, in writes
None, options)
File "/home/snow/venv/lib/python3.6/site-packages/hdf5storage/lowlevel.py", line 114, in write_data
m.write(f, grp, name, data, type_string, options)
File "/home/snow/venv/lib/python3.6/site-packages/hdf5storage/Marshallers.py", line 834, in write
**filters)
File "/home/snow/venv/lib/python3.6/site-packages/h5py/_hl/group.py", line 148, in create_dataset
dsid = dataset.make_new_dset(group, shape, dtype, data, name, **kwds)
File "/home/snow/venv/lib/python3.6/site-packages/h5py/_hl/dataset.py", line 137, in make_new_dset
dset_id = h5d.create(parent.id, name, tid, sid, dcpl=dcpl)
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/h5d.pyx", line 87, in h5py.h5d.create
ValueError: Unable to create dataset (no write intent on file)

ALL my clean and noisy train/test data are put in ./data.
hdf5storage=0.1.15
h5py==3.1.0
Could you help me solve it?

Question

If I want to change clean/noisy_trainset/testset_wav. What should I need to notice?

real-time

Hello. congrats on the awesome work. I was just wondering whether we can use the model to do real time noise cancellation?

error

I'm running "run_isegan.py" but it appears error "Error polling for event status: failed to query event: CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered".
My tensorflow is 1.13.1, cuda is 10.0.1, cudnn is 7.6.5 and keras is 2.2.4.

KeyError: 'Gtanh'

Hi @deepakbaby

Running the following command I encountered an error:

python run_isegan.py 

The error is:

Traceback (most recent call last):
File "run_isegan.py", line 117, in
G = generator(opts)
File "/home/betegon/isegan/models.py", line 120, in generator
if opts['Gtanh']:
KeyError: 'Gtanh'

It comes from the following line of code in models.py:

 # Add tanh of G uses tanh activation
if opts['Gtanh']:
    dec_out = Activation('tanh')(dec_out)   

The error is pretty easy to understand, there is no key called Gtanh in the dictionary opts, passed as argument when calling Generator(opts) from run_isegan.py. So it will be as easy as adding it as a Boolean in run_isegan.py, i.e.,:

opts['Gtanh'] = True # or false if better.

So my question is:

Is it better to use tanh activation or not?

Just let me know if you want me to make a pull request (but first you should add a LICENSE to the repository).

thanks a lot,

kind regards!

Pretrained Weights

Hi @deepakbaby

I just reviewed your paper (congrats for that, so good!) and I will like to give it a try.

Could you provide pre-trained weights of any configuration? This way people could check the functionality as the time consumed denoising files.

also, could you provide an example audio?
meaning a noisy audio and the cleaned audio, as a sanity check after installing all dependencies, so people can try the network and be sure everything work as expected.

Thank you for your time and effort,

regards!

Problem

I'm running "run_isegan.py" but it appears error "Error polling for event status: failed to query event: CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered".
My tensorflow is 1.13.1, cuda is 10.0.1, cudnn is 7.6.5 and keras is 2.2.4.