Add BNR2 model documentation

docs/source/audio/api.rst

@@ -38,6 +38,11 @@ Processing Models
     :members:
     :exclude-members: setup_training_data, setup_validation_data, training_step, on_validation_epoch_end, validation_step, setup_test_data, on_train_epoch_start
 
+.. autoclass:: nemo.collections.audio.models.BNR2
+    :show-inheritance:
+    :members:
+    :exclude-members: setup_training_data, setup_validation_data, training_step, on_validation_epoch_end, validation_step, setup_test_data, on_train_epoch_start
+
 
 Modules
 -------

docs/source/audio/models.rst

@@ -37,6 +37,11 @@ Flow Matching Model
 Flow matching model is a generative model using a noise-to-data process to transform the input (degraded) audio signal into the target (clean) audio signal :cite:`audio-models-ku2024generative`. The model consists of an encoder and decoder, a neural estimator, a flow model and a sampler.
 
 
+Background Noise Removal (BNR) Model
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+Background Noise Removal (BNR) 2.0 is a single-channel speech denoising model that uses the SEASR architecture. It combines convolutional feature extraction with GRU-based temporal modeling and a learnable masking mechanism operating in a learned transform domain. See `SEASR: A Speech Enhancement Model Using the SAEV Representation <https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10837982>`_ for details.
+
+
 References
 ----------
 

nemo/collections/audio/models/__init__.py

@@ -20,9 +20,11 @@
     SchroedingerBridgeAudioToAudioModel,
     ScoreBasedGenerativeAudioToAudioModel,
 )
+from nemo.collections.audio.models.maxine.bnr import BNR2
 
 __all__ = [
     "AudioToAudioModel",
+    "BNR2",
     "EncMaskDecAudioToAudioModel",
     "FlowMatchingAudioToAudioModel",
     "PredictiveAudioToAudioModel",

nemo/collections/audio/models/maxine/bnr.py

@@ -154,7 +154,20 @@ def _remove_weight_norm(m):
 
 
 class BNR2(AudioToAudioModel):
-    """Implementation of the BNR 2 model"""
+    """Maxine Background Noise Removal (BNR) 2.0 model.
+
+    BNR 2.0 is a single-channel speech denoising model that removes background
+    noise from audio to improve speech intelligibility and downstream ASR accuracy.
+    It uses the SEASR architecture, which combines convolutional feature extraction
+    with GRU-based temporal modeling and a learnable masking mechanism operating
+    in a learned transform domain.
+
+    The model operates at 16 kHz sample rate.
+
+    Reference:
+        `SEASR: A Speech Enhancement Model Using the SAEV Representation
+        <https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10837982>`_
+    """
 
     def __init__(self, cfg: DictConfig, trainer: Trainer = None):
         self.world_size = 1