UnicodeDecodeError: 'utf-8' codec can't decode byte 0xa4 in position 6: invalid start byte. about librimix

[JorisCos]

Hello,
Thanks for pointing out this issue.
The problem comes from the library pysndfx.
I believe you are using Windows right ?
I had trouble using this library on my Windows machine as well but it works just fine on Linux.
I will investigate further and come back to you with more information.

[JorisCos]

I couldn't replicate the error on my machine.
Have you installed sox on your machine ?
On a Windows environment with anaconda running conda install -c groakat sox
in your working environment should do the trick.
If it is still not working can you try to run this in a python console

from pysndfx import AudioEffectsChain    
import numpy as np    
s =np.random.randn(1,16000)    
fx = (AudioEffectsChain().speed(0.8))    
fx(s)

[cloverhsiao]

Yes, I am using Window.

conda install -c groakat sox printf()
solve the error caused by the library pysndfx.

However, I got another error like this.

And I tried to run this in a Linux VirtualBox environment.
The error reason is the same Index Error.

so I tried to modify the line in scripts/augment_train_noise.py
# Get 1st channel
# s = s[ : , 0 ]
change to
# Get 1st channel
# s = s

the running outcome will become like this.

in Linux VirtualBox environment

[JorisCos]

You have tried to run the augmentation on wham noise multiples times.
There are 20000 noises in wham_noise/tr before augmentation and 60000 after augmentation.
The 20000 original noise have 2 channels the 40000 augmented have one this is why you have this error.
Just skip augmentation IMO

[cloverhsiao]

Yes. Thank you so much.
I commented out the augment noise and successfully generate the dataset.

And I find out that the program will fail if you only want to generate types mix_clean.

I do this at first then I always got error message.

Then I change the types to only mix_both or mix_clean + mix both and successfully generate the dataset.

Thank you

[JorisCos]

Glad to hear that !
I will look into the last error thank you for reporting it ! 🚀