Use thread-safe loading of mel matrix IN audio.py

[mzamini92]

The mel_filters function is using a np.load function to load a pre-computed mel filterbank matrix. This function is not thread-safe, which means that if it is called from multiple threads at the same time, it may corrupt the data.

To fix this, you can use the torch.load function instead. This function is thread-safe, so it will not corrupt the data if it is called from multiple threads at the same time.

[jstoone]

Hi @mzamini92 this is great! I would suggest that you update the PR title, so that it reflects what the overall change strives to address.

Something like:

Use thread-safe loading of mel matrix

[mzamini92]

Hi @mzamini92 this is great! I would suggest that you update the PR title, so that it reflects what the overall change strives to address.

Something like:

Use thread-safe loading of mel matrix

Thanks for your comment. Done.

[jongwook]

Hi, thanks for pointing this out. Is there a documentation stating that np.load is not thread-safe? I think using torch.load() is a better idea in either case, but the docstring and the .npz file should be updated in conjunction.

[mzamini92]

Hi, thanks for pointing this out. Is there a documentation stating that np.load is not thread-safe? I think using torch.load() is a better idea in either case, but the docstring and the .npz file should be updated in conjunction.

Hi @jongwook . yes
https://numpy.org/devdocs/reference/generated/numpy.load.html
In the warning it says: Loading files that contain object arrays uses the pickle module, which is not secure against erroneous or maliciously constructed data. Consider passing allow_pickle=False to load data that is known not to contain object arrays for the safer handling of untrusted sources.
also, updated the docstring below.

[jongwook]

The warning is not related to thread safety, and torch.load() also has the same security hazard regarding unpickling. Also, torch.save does not perform compression unlike np.savez_compressed, which would add a huge binary to the git repo or make it slightly more annoying having to call a decompress function. I'll just add allow_pickle=False instead to address your concern.