Resnet 4 speaker verification

[mrouvier]

Hi, this PR for adding ResNet models and recipes for speaker verification. Be careful this recipe is not tested. A major problem of this recipe is that training Resnet model is time-intensive. I think for improvement the recipe it would be great to not calculate the acoustic feature on the fly but extract and save the acoustic feature in a file (once time for the whole corpus).

[anautsch]

Hi @mrouvier thank you for contributing this PR!

Can you please fetch the latest version of the develop branch into this PR? There has been an update on dependencies to avoid a github workflow testing error on the PLDA module (and github uses the PR branch, not the target branch to run tests).

There were tests failing on July 1st - I just re-report here what they were (in case they are unavailable to you) - this PR is stalling for a few months now w/o communication (sorry for this).

Before doc, unit & integration tests were executed, these two failed:

tests/consistency/test_docstrings.py F                                   [ 16%]
tests/consistency/test_recipe.py F...                                    [ 83%]

copy/paste:d from the github error log

	ERROR: The function forward in ./speechbrain/lobes/models/ResNet.py has no docstring. Please write it. For more info, see tests/consistency/DOCSTRINGS.md
	ERROR: The function forward in ./speechbrain/lobes/models/ResNet.py has no docstring. Please write it. For more info, see tests/consistency/DOCSTRINGS.md
	ERROR: The function forward in ./speechbrain/lobes/models/ResNet.py has no docstring. Please write it. For more info, see tests/consistency/DOCSTRINGS.md
	ERROR: The function forward in ./speechbrain/lobes/models/ResNet.py has no docstring. Please write it. For more info, see tests/consistency/DOCSTRINGS.md
	ERROR: The function forward in ./speechbrain/lobes/models/ResNet.py has no docstring. Please write it. For more info, see tests/consistency/DOCSTRINGS.md

and

tests/consistency/test_recipe.py:59: AssertionError
----------------------------- Captured stdout call -----------------------------
	ERROR: The file recipes/VoxCeleb/SpeakerRec/hparams/train_resnet.yaml is not listed in tests/recipes.csv. Please add it.                 For more info see tests/consistency/README.md

Was there progress on your end regarding the model testing?

As you mentioned, batch-independent computations are costly. There is a plan to get this into SpeechBrain - with the next major version. What's your take on how to proceed best?

[mravanelli]

@mrouvier, could you pelase take a look at the errrors mentioend by @anautsch?

[Adel-Moumen]

Merged in #2090. Thanks @mrouvier for your contribution! :)