nvidia-nemo--speech/tests/collections/audio/test_audio_models_schroedinger_bridge.py

# Copyright (c) 2025, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import itertools
import json

import einops
import lhotse
import lightning.pytorch as pl
import numpy as np
import pytest
import soundfile as sf
import torch
from omegaconf import DictConfig

from nemo.collections.audio.models.enhancement import SchroedingerBridgeAudioToAudioModel


@pytest.fixture(params=["nemo_manifest", "lhotse_cuts"])
def mock_dataset_config(tmp_path, request):
    num_files = 8
    num_samples = 16000

    for i in range(num_files):
        data = np.random.randn(num_samples, 1)
        sf.write(tmp_path / f"audio_{i}.wav", data, 16000)

    if request.param == "lhotse_cuts":
        with lhotse.CutSet.open_writer(tmp_path / "cuts.jsonl") as writer:
            for i in range(num_files):
                recording = lhotse.Recording.from_file(tmp_path / f"audio_{i}.wav")
                cut = lhotse.MonoCut(
                    id=f"audio_{i}",
                    start=0,
                    channel=0,
                    duration=num_samples / 16000,
                    recording=recording,
                    custom={"target_recording": recording},
                )
                writer.write(cut)

            return {
                'cuts_path': str(tmp_path / "cuts.jsonl"),
                'use_lhotse': True,
                'batch_size': 2,
                'num_workers': 1,
            }
    elif request.param == "nemo_manifest":
        with (tmp_path / "small_manifest.jsonl").open("w") as f:
            for i in range(num_files):
                entry = {
                    "noisy_filepath": str(tmp_path / f"audio_{i}.wav"),
                    "clean_filepath": str(tmp_path / f"audio_{i}.wav"),
                    "duration": num_samples / 16000,
                    "offset": 0,
                }
                f.write(f"{json.dumps(entry)}\n")
        return {
            'manifest_filepath': str(tmp_path / "small_manifest.jsonl"),
            'input_key': 'noisy_filepath',
            'target_key': 'clean_filepath',
            'use_lhotse': False,
            'batch_size': 2,
            'num_workers': 1,
        }
    else:
        raise NotImplementedError(f"Dataset type {request.param} not implemented")


@pytest.fixture()
def schroedinger_bridge_model_ncsn_params():

    model = {
        'sample_rate': 16000,
        'num_outputs': 1,
        'normalize_input': True,
        'max_utts_evaluation_metrics': 50,
    }
    encoder = {
        '_target_': 'nemo.collections.audio.modules.transforms.AudioToSpectrogram',
        'fft_length': 510,
        'hop_length': 128,
        'magnitude_power': 0.5,
        'scale': 0.33,
    }
    decoder = {
        '_target_': 'nemo.collections.audio.modules.transforms.SpectrogramToAudio',
        'fft_length': encoder['fft_length'],
        'hop_length': encoder['hop_length'],
        'magnitude_power': encoder['magnitude_power'],
        'scale': encoder['scale'],
    }
    estimator = {
        '_target_': 'nemo.collections.audio.parts.submodules.ncsnpp.SpectrogramNoiseConditionalScoreNetworkPlusPlus',
        'in_channels': 2,  # single-channel noisy input
        'out_channels': 1,  # single-channel estimate
        'conditioned_on_time': True,
        'channels': [8, 8, 8, 8, 8],
        'num_res_blocks': 3,  # increased number of res blocks
        'pad_time_to': 64,  # pad to 64 frames for the time dimension
        'pad_dimension_to': 0,  # no padding in the frequency dimension
    }

    loss_encoded = {
        '_target_': 'nemo.collections.audio.losses.audio.MSELoss',
        'ndim': 4,
    }  # computed in the time domain

    loss_time = {'_target_': 'nemo.collections.audio.losses.audio.MAELoss'}

    noise_schedule = {
        '_target_': 'nemo.collections.audio.parts.submodules.schroedinger_bridge.SBNoiseScheduleVE',
        'k': 2.6,
        'c': 0.4,
        'time_min': 1e-4,
        'time_max': 1.0,
        'num_steps': 1000,  # num steps for the forward process
    }

    sampler = {
        '_target_': 'nemo.collections.audio.parts.submodules.schroedinger_bridge.SBSampler',
        'time_min': 1e-4,
        'time_max': 1.0,
        'num_steps': 5,  # num steps for the reverse process
    }

    metrics = {
        'val': {
            'sisdr': {
                '_target_': 'torchmetrics.audio.ScaleInvariantSignalDistortionRatio',
            },
        },
    }

    model_config = DictConfig(
        {
            'sample_rate': model['sample_rate'],
            'num_outputs': model['num_outputs'],
            'normalize_input': model['normalize_input'],
            'max_utts_evaluation_metrics': model['max_utts_evaluation_metrics'],
            'encoder': DictConfig(encoder),
            'decoder': DictConfig(decoder),
            'estimator': DictConfig(estimator),
            'loss_encoded': DictConfig(loss_encoded),
            'loss_time': DictConfig(loss_time),
            'loss_time_weight': 0.001,
            'estimator_output': 'data_prediction',
            'noise_schedule': DictConfig(noise_schedule),
            'sampler': DictConfig(sampler),
            'metrics': metrics,
            'optim': {
                'name': 'adam',
                'lr': 0.001,
                'betas': (0.9, 0.98),
            },
        }
    )

    return model_config


@pytest.fixture()
def schroedinger_bridge_model_ncsn(schroedinger_bridge_model_ncsn_params):
    with torch.random.fork_rng():
        torch.random.manual_seed(0)
        model = SchroedingerBridgeAudioToAudioModel(cfg=schroedinger_bridge_model_ncsn_params)
    return model


@pytest.fixture()
def schroedinger_bridge_model_ncsn_with_trainer_and_mock_dataset(
    schroedinger_bridge_model_ncsn_params, mock_dataset_config
):
    # Add train and validation dataset configs
    schroedinger_bridge_model_ncsn_params["train_ds"] = {**mock_dataset_config, "shuffle": True}
    schroedinger_bridge_model_ncsn_params["validation_ds"] = {**mock_dataset_config, "shuffle": False}

    # Trainer config
    trainer_cfg = {
        "max_epochs": -1,
        "max_steps": 8,
        "logger": False,
        "use_distributed_sampler": False,
        "val_check_interval": 2,
        "limit_train_batches": 4,
        "accelerator": "cpu",
        "enable_checkpointing": False,
    }
    schroedinger_bridge_model_ncsn_params["trainer"] = trainer_cfg

    trainer = pl.Trainer(**trainer_cfg)

    with torch.random.fork_rng():
        torch.random.manual_seed(0)
        model = SchroedingerBridgeAudioToAudioModel(cfg=schroedinger_bridge_model_ncsn_params, trainer=trainer)

    return model, trainer


class TestSchroedingerBridgeModelNCSN:
    """Test Schroedinger Bridge model with NCSN estimator."""

    @pytest.mark.unit
    def test_constructor(self, schroedinger_bridge_model_ncsn):
        """Test that the model can be constructed from a config dict."""
        model = schroedinger_bridge_model_ncsn.train()
        confdict = model.to_config_dict()
        instance2 = SchroedingerBridgeAudioToAudioModel.from_config_dict(confdict)
        assert isinstance(instance2, SchroedingerBridgeAudioToAudioModel)

    @pytest.mark.unit
    @pytest.mark.parametrize(
        "batch_size, sample_len",
        [
            (4, 4),  # Example 1
            (2, 8),  # Example 2
            (1, 10),  # Example 3
        ],
    )
    def test_forward_infer(self, schroedinger_bridge_model_ncsn, batch_size, sample_len):
        """Test that the model can run forward inference."""
        model = schroedinger_bridge_model_ncsn.eval()
        confdict = model.to_config_dict()
        sampling_rate = confdict['sample_rate']
        rng = torch.Generator()
        rng.manual_seed(0)
        input_signal = torch.randn(size=(batch_size, 1, sample_len * sampling_rate), generator=rng)
        input_signal_length = (sample_len * sampling_rate) * torch.ones(batch_size, dtype=torch.int)

        with torch.no_grad():
            # batch size 1
            output_list = []
            output_length_list = []
            for i in range(input_signal.size(0)):
                output, output_length = model.forward(
                    input_signal=input_signal[i : i + 1], input_length=input_signal_length[i : i + 1]
                )
                output_list.append(output)
                output_length_list.append(output_length)
            output_instance = torch.cat(output_list, 0)
            output_length_instance = torch.cat(output_length_list, 0)

            # batch size batch_size
            output_batch, output_length_batch = model.forward(
                input_signal=input_signal, input_length=input_signal_length
            )

        # It is generative model so we do not check the diffenence between output_instance and output_batch

        # Check that the output and output length are the same for the instance and batch
        assert output_instance.shape == output_batch.shape
        assert output_length_instance.shape == output_length_batch.shape

    def test_training_step(self, schroedinger_bridge_model_ncsn_with_trainer_and_mock_dataset):
        model, _ = schroedinger_bridge_model_ncsn_with_trainer_and_mock_dataset
        model = model.train()
        # _step calls self.log for component losses, which requires a Lightning loop context.
        # Disable logging since we're calling _step directly outside the training loop.
        model.log = lambda *args, **kwargs: None

        for batch in itertools.islice(model._train_dl, 2):
            input_signal, target_signal, input_length = model._parse_batch(batch)

            loss = model._step(target_signal=target_signal, input_signal=input_signal, input_length=input_length)
            loss.backward()

    def test_model_training(self, schroedinger_bridge_model_ncsn_with_trainer_and_mock_dataset):
        """
        Test that the model can be trained for a few steps. An evaluation step is also expected.
        """
        model, trainer = schroedinger_bridge_model_ncsn_with_trainer_and_mock_dataset
        model = model.train()

        trainer.fit(model)