[ASR] GSS-based mask estimator

[anteju]

What does this PR do ?

Adding GSS mask estimator for use in multispeaker scenarios.

Collection: ASR

Changelog

• Added an implementation of mask estimator using directional statistics clustering and activity guidance
• Added a unit test to test the module can be initialized and the output has the expected shape

Usage

import torch
from nemo.collections.asr.modules.audio_modules import MaskEstimatorGSS

batch_size, num_channels, num_subbands, num_frames = 1, 4, 257, 100
num_outputs = 3

# input (mixture) spectrogram
spec = torch.randn(batch_size, num_channels, num_subbands, num_frames, dtype=torch.cfloat)
# estimated source activity
source_activity = torch.randn(batch_size, num_outputs, num_frames) > 0.5
# mask estimator
me = MaskEstimatorGSS(num_iterations=5)
# estimated source masks
masks = me(input=spec, activity=source_activity)

assert masks.shape == (batch_size, num_outputs, num_subbands, num_frames)

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

• Related to # (issue)

[anteju]

jenkins

[anteju]

jenkins

[anteju]

jenkins

[tango4j]

I ran the example code in the PR description without a problem,
I ran tests/collections/asr/test_audio_modules.py without a problem for both cpu and gpu.
I think the test runs can be faster if we use pytest.mark.parameterize() for TestMaskEstimator. This is the only major comment I would like to make.

[tango4j]

Seems like the comments are well addressed. The new test is passing without a problem.
Approving.

[tango4j]

ImportError while loading conftest '/home/taejinp/projects/gss_mask/NeMo/tests/conftest.py'. tests/conftest.py:28: in <module> from tests.fixtures.tts import * tests/fixtures/tts.py:21: in <module> from nemo.collections.asr.parts.utils.manifest_utils import read_manifest nemo/collections/asr/__init__.py:15: in <module> from nemo.collections.asr import data, losses, models, modules nemo/collections/asr/models/__init__.py:17: in <module> from nemo.collections.asr.models.classification_models import EncDecClassificationModel, EncDecFrameClassificationModel nemo/collections/asr/models/classification_models.py:29: in <module> from nemo.collections.asr.data import audio_to_label_dataset, feature_to_label_dataset nemo/collections/asr/data/audio_to_label_dataset.py:19: in <module> from nemo.collections.asr.data.audio_to_text_dataset import convert_to_config_list, get_chain_dataset nemo/collections/asr/data/audio_to_text_dataset.py:28: in <module> from nemo.collections.asr.data.huggingface.hf_audio_to_text_dataset import ( nemo/collections/asr/data/huggingface/hf_audio_to_text_dataset.py:17: in <module> from nemo.collections.asr.data.huggingface.hf_audio_to_text import ( nemo/collections/asr/data/huggingface/hf_audio_to_text.py:34: in <module> class HFTextProcessor: nemo/collections/asr/data/huggingface/hf_audio_to_text.py:55: in HFTextProcessor symbols_to_keep: Optional[str | List[str]] = None, E TypeError: unsupported operand type(s) for |: 'type' and '_GenericAlias'

This error keeps appearing regardless of commits so commented when I was running the tests.
This might take some time to pass the test; Note that this error is nothing to do with this (#7849) PR.

[anteju]

jenkins