dec duration

:

.flake8

@@ -1,5 +1,5 @@
 [flake8]
-per-file-ignores = "mayavoz/model/__init__.py:F401"
+per-file-ignores = __init__.py:F401
 ignore = E203, E266, E501, W503
 # line length is intentionally set to 80 here because black uses Bugbear
 # See https://github.com/psf/black/blob/master/README.md#line-length for more details

.gitattributes

@@ -1 +0,0 @@
-notebooks/** linguist-vendored

.github/workflows/ci.yaml

@@ -1,13 +1,13 @@
 # This workflow will install Python dependencies, run tests and lint with a variety of Python versions
 # For more information see: https://help.github.com/actions/language-and-framework-guides/using-python-with-github-actions
 
-name: mayavoz
+name: Enhancer
 
 on:
   push:
-    branches: [ main ]
+    branches: [ dev ]
   pull_request:
-    branches: [ main ]
+    branches: [ dev ]
 jobs:
   build:
     runs-on: ubuntu-latest
@@ -40,12 +40,12 @@ jobs:
         sudo apt-get install libsndfile1
         pip install -r requirements.txt
         pip install black pytest-cov
-    - name: Install mayavoz
+    - name: Install enhancer
       run: |
           pip install -e .[dev,testing]
     - name: Run black
       run:
-        black --check .  --exclude mayavoz/version.py
+        black --check .  --exclude enhancer/version.py
     - name: Test with pytest
       run:
-        pytest tests --cov=mayavoz/
+        pytest tests --cov=enhancer/

.gitignore

@@ -1,8 +1,4 @@
 #local
-cleaned_my_voice.wav
-lightning_logs/
-my_voice.wav
-pretrained/
 *.ckpt
 *_local.yaml
 cli/train_config/dataset/Vctk_local.yaml

.pre-commit-config.yaml

@@ -23,7 +23,6 @@ repos:
       hooks:
       - id: flake8
         args: ['--ignore=E203,E501,F811,E712,W503']
-        exclude: __init__.py
 
     # Formatting, Whitespace, etc
     - repo: https://github.com/pre-commit/pre-commit-hooks

CONTRIBUTING.md

@@ -1,46 +0,0 @@
-# Contributing
-
-Hi there 👋
-
-If you're reading this I hope that you're looking forward to adding value to Mayavoz. This document will help you to get started with your journey.
-
-## How to get your code in Mayavoz
-
-1. We use git and GitHub.
-
-2. Fork the mayavoz repository (https://github.com/shahules786/mayavoz) on GitHub under your own account. (This creates a copy of mayavoz under your account, and GitHub knows where it came from, and we typically call this “upstream”.)
-
-3. Clone your own mayavoz repository. git clone https://github.com/ <your-account> /mayavoz (This downloads the git repository to your machine, git knows where it came from, and calls it “origin”.)
-
-4. Create a branch for each specific feature you are developing. git checkout -b your-branch-name
-
-5. Make + commit changes. git add files-you-changed ... git commit -m "Short message about what you did"
-
-6. Push the branch to your GitHub repository. git push origin your-branch-name
-
-7. Navigate to GitHub, and create a pull request from your branch to the upstream repository mayavoz/mayavoz, to the “develop” branch.
-
-8. The Pull Request (PR) appears on the upstream repository. Discuss your contribution there. If you push more changes to your branch on GitHub (on your repository), they are added to the PR.
-
-9. When the reviewer is satisfied that the code improves repository quality, they can merge.
-
-Note that CI tests will be run when you create a PR. If you want to be sure that your code will not fail these tests, we have set up pre-commit hooks that you can install.
-
-**If you're worried about things not being perfect with your code, we will work togethor and make it perfect. So, make your move!**
-
-## Formating
-
-We use [black](https://black.readthedocs.io/en/stable/) and [flake8](https://flake8.pycqa.org/en/latest/) for code formating. Please ensure that you use the same before submitting the PR.
-
-
-## Testing
-We adopt unit testing using [pytest](https://docs.pytest.org/en/latest/contents.html)
-Please make sure that adding your new component does not decrease test coverage.
-
-## Other tools
-The use of [per-commit](https://pre-commit.com/) is recommended to ensure different requirements such as code formating, etc.
-
-## How to start contributing to Mayavoz?
-
-1. Checkout issues marked as `good first issue`, let us know you're interested in working on some issue by commenting under it.
-2. For others, I would suggest you to explore mayavoz. One way to do is to use it to train your own model. This was you might end by finding a new unreported bug or getting an idea to improve Mayavoz.

LICENSE

@@ -1,20 +0,0 @@
-MIT License
-
-Copyright (c) 2022 Shahul Es
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.

MANIFEST.in

@@ -1,4 +0,0 @@
-recursive-include mayavoz *.py
-recursive-include mayavoz *.yaml
-global-exclude *.pyc
-global-exclude __pycache__

README.md

@@ -2,52 +2,24 @@
   <img src="https://user-images.githubusercontent.com/25312635/195514652-e4526cd1-1177-48e9-a80d-c8bfdb95d35f.png" />
 </p>
 
-![GitHub Workflow Status](https://img.shields.io/github/actions/workflow/status/shahules786/mayavoz/ci.yaml?branch=main)
+mayavoz is a Pytorch-based opensource toolkit for speech enhancement. It is designed to save time for audio researchers. Is provides easy to use pretrained audio enhancement models and facilitates highly customisable model training.
-![GitHub](https://img.shields.io/github/license/shahules786/enhancer)
-![GitHub issues](https://img.shields.io/github/issues/shahules786/enhancer?logo=GitHub)
-![GitHub Repo stars](https://img.shields.io/github/stars/shahules786/enhancer?style=social)
 
-mayavoz is a Pytorch-based opensource toolkit for speech enhancement. It is designed to save time for audio practioners & researchers. It provides easy to use pretrained speech enhancement models and facilitates highly customisable model training.
+| **[Quick Start]()** | **[Installation]()** | **[Tutorials]()** | **[Available Recipes]()**
-
-| **[Quick Start](#quick-start-fire)** | **[Installation](#installation)** | **[Tutorials](https://github.com/shahules786/enhancer/tree/main/notebooks)** | **[Available Recipes](#recipes)** | **[Demo](#demo)**
 ## Key features :key:
 
-* Various pretrained models nicely integrated with [huggingface hub](https://huggingface.co/docs/hub/index) :hugs: that users can select and use without any hastle.
+* Various pretrained models nicely integrated with huggingface 	:hugs: that users can select and use without any hastle.
-* :package: Ability to train and validate your own custom speech enhancement models with just under 10 lines of code!
+* :package: Ability to train and validation your own custom speech enhancement models with just under 10 lines of code!
 * :magic_wand: A command line tool that facilitates training of highly customisable speech enhacement models from the terminal itself!
-* :zap: Supports multi-gpu training integrated with [Pytorch Lightning](https://pytorchlightning.ai/).
+* :zap: Supports multi-gpu training integrated with Pytorch Lightning.
-* :shield: data augmentations integrated using [torch-augmentations](https://github.com/asteroid-team/torch-audiomentations)
-
-
-## Demo
-
-Noisy speech followed by enhanced version.
-
-https://user-images.githubusercontent.com/25312635/203756185-737557f4-6e21-4146-aa2c-95da69d0de4c.mp4
-
-
 
 ## Quick Start :fire:
 ``` python
-from mayavoz.models import Mayamodel
+from mayavoz import Mayamodel
 
-model = Mayamodel.from_pretrained("shahules786/mayavoz-waveunet-valentini-28spk")
+model = Mayamodel.from_pretrained("mayavoz/waveunet")
-model.enhance("noisy_audio.wav")
+model("noisy_audio.wav")
 ```
 
-## Recipes
-
-| Model     | Dataset      | STOI    | PESQ  | URL                           |
-| :---:     |  :---:       | :---:   | :---: | :---:                         |
-| WaveUnet  | Valentini-28spk   | 0.836   | 2.78  |  shahules786/mayavoz-waveunet-valentini-28spk      |
-| Demucs    | Valentini-28spk   | 0.961   | 2.56  |  shahules786/mayavoz-demucs-valentini-28spk       |
-| DCCRN     | Valentini-28spk   | 0.724   | 2.55  |  shahules786/mayavoz-dccrn-valentini-28spk         |
-| Demucs     | MS-SNSD-20hrs  | 0.56 | 1.26  | shahules786/mayavoz-demucs-ms-snsd-20       |
-
-Test scores are based on respective test set associated with train dataset.
-
-**See [tutorials](/notebooks/) to train your custom model**
-
 ## Installation
 Only Python 3.8+ is officially supported (though it might work with Python 3.7)
 
@@ -69,10 +41,3 @@ git clone url
 cd mayavoz
 pip install -e .
 ```
-
-## Support
-
-For commercial enquiries and scientific consulting, please [contact me](https://shahules786.github.io/).
-
-### Acknowledgements
-Sincere gratitude to [AMPLYFI](https://amplyfi.com/) for supporting this project.

enhancer/__init__.py

@@ -1,2 +1 @@
 __import__("pkg_resources").declare_namespace(__name__)
-from mayavoz.models import Mayamodel

enhancer/cli/train.py

@@ -60,9 +60,9 @@ def main(config: DictConfig):
 
     if parameters.get("Early_stop", False):
         early_stopping = EarlyStopping(
-            monitor="val_loss",
+            monitor=f"valid_{parameters.get('EarlyStopping_metric','loss')}",
             mode=direction,
-            min_delta=0.0,
+            min_delta=parameters.get("EarlyStopping_delta", 0.00),
             patience=parameters.get("EarlyStopping_patience", 10),
             strict=True,
             verbose=False,
@@ -93,7 +93,7 @@ def main(config: DictConfig):
 
     trainer = instantiate(config.trainer, logger=logger, callbacks=callbacks)
     trainer.fit(model)
-    trainer.test(model)
+    trainer.test(ckpt_path="best")
 
     logger.experiment.log_artifact(
         logger.run_id, f"{trainer.default_root_dir}/config_log.yaml"

enhancer/cli/train_config/config.yaml

@@ -1,6 +1,6 @@
 defaults:
-  - model : Demucs
+  - model : DCCRN
-  - dataset : Vctk
+  - dataset : DNS-2020
   - optimizer : Adam
   - hyperparameters : default
   - trainer : default

enhancer/cli/train_config/dataset/DNS-2020.yaml

@@ -0,0 +1,13 @@
+_target_: enhancer.data.dataset.EnhancerDataset
+root_dir : /scratch/c.sistc3/MS-SNSD/DNS20
+name : dns-2020
+duration : 1.5
+stride : 1
+sampling_rate: 16000
+batch_size: 32
+min_valid_minutes: 25.0
+files:
+  train_clean : CleanSpeech_training
+  test_clean : CleanSpeech_testing
+  train_noisy : NoisySpeech_training
+  test_noisy : NoisySpeech_testing

enhancer/cli/train_config/dataset/Vctk.yaml

@@ -1,8 +1,8 @@
-_target_: mayavoz.data.dataset.MayaDataset
+_target_: enhancer.data.dataset.EnhancerDataset
 name : vctk
 root_dir : /scratch/c.sistc3/DS_10283_2791
-duration : 4.5
+duration : 2
-stride : 0.5
+stride : 1
 sampling_rate: 16000
 batch_size: 32
 min_valid_minutes : 25

enhancer/cli/train_config/dataset/Vctk_local.yaml

@@ -0,0 +1,13 @@
+_target_: enhancer.data.dataset.EnhancerDataset
+name : vctk
+root_dir : /Users/shahules/Myprojects/enhancer/datasets/vctk
+duration : 1.0
+sampling_rate: 16000
+batch_size: 64
+num_workers : 0
+
+files:
+  train_clean : clean_testset_wav
+  test_clean : clean_testset_wav
+  train_noisy : noisy_testset_wav
+  test_noisy : noisy_testset_wav

enhancer/cli/train_config/hyperparameters/default.yaml

@@ -2,6 +2,7 @@ loss : si-snr
 metric : [stoi,pesq]
 lr : 0.001
 ReduceLr_patience : 10
+Early_stop : False
 ReduceLr_factor : 0.5
-min_lr : 0.000001
+min_lr : 0.0000001
-EarlyStopping_factor : 10
+EarlyStopping_patience : 10

enhancer/cli/train_config/mlflow/experiment.yaml

@@ -0,0 +1,2 @@
+experiment_name : shahules/enhancer
+run_name : dccrn-dns20

enhancer/cli/train_config/model/DCCRN.yaml

@@ -1,10 +1,10 @@
-_target_: mayavoz.models.dccrn.DCCRN
+_target_: enhancer.models.dccrn.DCCRN
 num_channels: 1
 sampling_rate : 16000
 complex_lstm : True
 complex_norm : True
 complex_relu : True
-masking_mode : True
+masking_mode : "E"
 
 encoder_decoder:
   initial_output_channels : 32

enhancer/cli/train_config/model/Demucs.yaml

@@ -1,4 +1,4 @@
-_target_: mayavoz.models.demucs.Demucs
+_target_: enhancer.models.demucs.Demucs
 num_channels: 1
 resample: 4
 sampling_rate : 16000

enhancer/cli/train_config/model/WaveUnet.yaml

@@ -1,4 +1,4 @@
-_target_: mayavoz.models.waveunet.WaveUnet
+_target_: enhancer.models.waveunet.WaveUnet
 num_channels : 1
 depth : 9
 initial_output_channels: 24

enhancer/cli/train_config/trainer/default.yaml

@@ -2,7 +2,7 @@ _target_: pytorch_lightning.Trainer
 accelerator: gpu
 accumulate_grad_batches: 1
 amp_backend: native
-auto_lr_find: True
+auto_lr_find: False
 auto_scale_batch_size: False
 auto_select_gpus: True
 benchmark: False
@@ -23,7 +23,7 @@ limit_test_batches: 1.0
 limit_train_batches: 1.0
 limit_val_batches: 1.0
 log_every_n_steps: 50
-max_epochs: 200
+max_epochs: 100
 max_steps: -1
 max_time: null
 min_epochs: 1
@@ -43,4 +43,3 @@ sync_batchnorm: False
 tpu_cores: null
 track_grad_norm: -1
 val_check_interval: 1.0
-weights_save_path: null

enhancer/data/__init__.py

@@ -0,0 +1 @@
+from enhancer.data.dataset import EnhancerDataset

enhancer/data/dataset.py

@@ -1,8 +1,6 @@
 import math
 import multiprocessing
 import os
-import sys
-import warnings
 from pathlib import Path
 from typing import Optional
 
@@ -13,11 +11,11 @@ import torch.nn.functional as F
 from torch.utils.data import DataLoader, Dataset, RandomSampler
 from torch_audiomentations import Compose
 
-from mayavoz.data.fileprocessor import Fileprocessor
+from enhancer.data.fileprocessor import Fileprocessor
-from mayavoz.utils import check_files
+from enhancer.utils import check_files
-from mayavoz.utils.config import Files
+from enhancer.utils.config import Files
-from mayavoz.utils.io import Audio
+from enhancer.utils.io import Audio
-from mayavoz.utils.random import create_unique_rng
+from enhancer.utils.random import create_unique_rng
 
 LARGE_NUM = 2147483647
 
@@ -82,21 +80,6 @@ class TaskDataset(pl.LightningDataModule):
         self._validation = []
         if num_workers is None:
             num_workers = multiprocessing.cpu_count() // 2
-        if num_workers is None:
-            num_workers = multiprocessing.cpu_count() // 2
-
-        if (
-            num_workers > 0
-            and sys.platform == "darwin"
-            and sys.version_info[0] >= 3
-            and sys.version_info[1] >= 8
-        ):
-            warnings.warn(
-                "num_workers > 0 is not supported with macOS and Python 3.8+: "
-                "setting num_workers = 0."
-            )
-            num_workers = 0
-
         self.num_workers = num_workers
         if min_valid_minutes > 0.0:
             self.min_valid_minutes = min_valid_minutes
@@ -152,7 +135,7 @@ class TaskDataset(pl.LightningDataModule):
             speaker_index = rng.choice(possible_indices)
             possible_indices.remove(speaker_index)
             speaker_name = all_speakers[speaker_index]
-            print(f"Selected f{speaker_name} for valid")
+            print(f"Selected {speaker_name} for valid")
             file_indices = [
                 i
                 for i, file in enumerate(data)
@@ -265,7 +248,7 @@ class TaskDataset(pl.LightningDataModule):
         )
 
 
-class MayaDataset(TaskDataset):
+class EnhancerDataset(TaskDataset):
     """
     Dataset object for creating clean-noisy speech enhancement datasets
     paramters:
@@ -275,7 +258,7 @@ class MayaDataset(TaskDataset):
         root directory of the dataset containing clean/noisy folders
     files : Files
         dataclass containing train_clean, train_noisy, test_clean, test_noisy
-        folder names (refer mayavoz.utils.Files dataclass)
+        folder names (refer enhancer.utils.Files dataclass)
     min_valid_minutes: float
         minimum validation split size time in minutes
         algorithm randomly select n speakers (>=min_valid_minutes) from train data to form validation data.

enhancer/data/fileprocessor.py

@@ -93,9 +93,9 @@ class Fileprocessor:
     def from_name(cls, name: str, clean_dir, noisy_dir, matching_function=None):
 
         if matching_function is None:
-            if name.lower() in ("vctk", "valentini"):
+            if name.lower() == "vctk":
                 return cls(clean_dir, noisy_dir, ProcessorFunctions.one_to_one)
-            elif name.lower() == "ms-snsd":
+            elif name.lower() == "dns-2020":
                 return cls(clean_dir, noisy_dir, ProcessorFunctions.one_to_many)
             else:
                 raise ValueError(

enhancer/inference.py

@@ -8,7 +8,7 @@ from librosa import load as load_audio
 from scipy.io import wavfile
 from scipy.signal import get_window
 
-from mayavoz.utils import Audio
+from enhancer.utils import Audio
 
 
 class Inference:
@@ -95,7 +95,6 @@ class Inference:
     ):
         """
         stitch batched waveform into single waveform. (Overlap-add)
-        inspired from https://github.com/asteroid-team/asteroid
         arguments:
             data: batched waveform
             window_size : window_size used to batch waveform

enhancer/loss.py

@@ -1,4 +1,4 @@
-import warnings
+import logging
 
 import numpy as np
 import torch
@@ -134,7 +134,7 @@ class Pesq:
             try:
                 pesq_values.append(self.pesq(pred.squeeze(), target_.squeeze()))
             except Exception as e:
-                warnings.warn(f"{e} error occured while calculating PESQ")
+                logging.warning(f"{e} error occured while calculating PESQ")
         return torch.tensor(np.mean(pesq_values))
 
 

enhancer/models/__init__.py

@@ -0,0 +1,3 @@
+from enhancer.models.demucs import Demucs
+from enhancer.models.model import Model
+from enhancer.models.waveunet import WaveUnet

enhancer/models/complexnn/__init__.py

@@ -0,0 +1,5 @@
+from enhancer.models.complexnn.conv import ComplexConv2d  # noqa
+from enhancer.models.complexnn.conv import ComplexConvTranspose2d  # noqa
+from enhancer.models.complexnn.rnn import ComplexLSTM  # noqa
+from enhancer.models.complexnn.utils import ComplexBatchNorm2D  # noqa
+from enhancer.models.complexnn.utils import ComplexRelu  # noqa

enhancer/models/complexnn/conv.py

@@ -129,7 +129,7 @@ class ComplexConvTranspose2d(nn.Module):
         imag_real = self.real_conv(imag)
 
         real = real_real - imag_imag
-        imag = real_imag + imag_real
+        imag = real_imag - imag_real
 
         out = torch.cat([real, imag], 1)
 

enhancer/models/dccrn.py

@@ -1,22 +1,22 @@
-import warnings
+import logging
 from typing import Any, List, Optional, Tuple, Union
 
 import torch
 import torch.nn.functional as F
 from torch import nn
 
-from mayavoz.data import MayaDataset
+from enhancer.data import EnhancerDataset
-from mayavoz.models import Mayamodel
+from enhancer.models import Model
-from mayavoz.models.complexnn import (
+from enhancer.models.complexnn import (
     ComplexBatchNorm2D,
     ComplexConv2d,
     ComplexConvTranspose2d,
     ComplexLSTM,
     ComplexRelu,
 )
-from mayavoz.models.complexnn.utils import complex_cat
+from enhancer.models.complexnn.utils import complex_cat
-from mayavoz.utils.transforms import ConviSTFT, ConvSTFT
+from enhancer.utils.transforms import ConviSTFT, ConvSTFT
-from mayavoz.utils.utils import merge_dict
+from enhancer.utils.utils import merge_dict
 
 
 class DCCRN_ENCODER(nn.Module):
@@ -98,7 +98,7 @@ class DCCRN_DECODER(nn.Module):
         return self.decoder(waveform)
 
 
-class DCCRN(Mayamodel):
+class DCCRN(Model):
 
     STFT_DEFAULTS = {
         "window_len": 400,
@@ -134,17 +134,17 @@ class DCCRN(Mayamodel):
         num_channels: int = 1,
         sampling_rate=16000,
         lr: float = 1e-3,
-        dataset: Optional[MayaDataset] = None,
+        dataset: Optional[EnhancerDataset] = None,
         duration: Optional[float] = None,
         loss: Union[str, List, Any] = "mse",
         metric: Union[str, List] = "mse",
     ):
         duration = (
-            dataset.duration if isinstance(dataset, MayaDataset) else duration
+            dataset.duration if isinstance(dataset, EnhancerDataset) else None
         )
         if dataset is not None:
             if sampling_rate != dataset.sampling_rate:
-                warnings.warn(
+                logging.warning(
                     f"model sampling rate {sampling_rate} should match dataset sampling rate {dataset.sampling_rate}"
                 )
                 sampling_rate = dataset.sampling_rate

enhancer/models/demucs.py

@@ -1,14 +1,14 @@
+import logging
 import math
-import warnings
 from typing import List, Optional, Union
 
 import torch.nn.functional as F
 from torch import nn
 
-from mayavoz.data.dataset import MayaDataset
+from enhancer.data.dataset import EnhancerDataset
-from mayavoz.models.model import Mayamodel
+from enhancer.models.model import Model
-from mayavoz.utils.io import Audio as audio
+from enhancer.utils.io import Audio as audio
-from mayavoz.utils.utils import merge_dict
+from enhancer.utils.utils import merge_dict
 
 
 class DemucsLSTM(nn.Module):
@@ -88,7 +88,7 @@ class DemucsDecoder(nn.Module):
         return out
 
 
-class Demucs(Mayamodel):
+class Demucs(Model):
     """
     Demucs model from https://arxiv.org/pdf/1911.13254.pdf
     parameters:
@@ -102,8 +102,8 @@ class Demucs(Mayamodel):
             sampling rate of input audio
         lr : float, defaults to 1e-3
             learning rate used for training
-        dataset: MayaDataset, optional
+        dataset: EnhancerDataset, optional
-            MayaDataset object containing train/validation data for training
+            EnhancerDataset object containing train/validation data for training
         duration : float, optional
             chunk duration in seconds
         loss : string or List of strings
@@ -135,18 +135,17 @@ class Demucs(Mayamodel):
         sampling_rate=16000,
         normalize=True,
         lr: float = 1e-3,
-        dataset: Optional[MayaDataset] = None,
+        dataset: Optional[EnhancerDataset] = None,
-        duration: Optional[float] = None,
         loss: Union[str, List] = "mse",
         metric: Union[str, List] = "mse",
         floor=1e-3,
     ):
         duration = (
-            dataset.duration if isinstance(dataset, MayaDataset) else duration
+            dataset.duration if isinstance(dataset, EnhancerDataset) else None
         )
         if dataset is not None:
             if sampling_rate != dataset.sampling_rate:
-                warnings.warn(
+                logging.warning(
                     f"model sampling rate {sampling_rate} should match dataset sampling rate {dataset.sampling_rate}"
                 )
                 sampling_rate = dataset.sampling_rate

enhancer/models/model.py

@@ -13,21 +13,17 @@ from pytorch_lightning.utilities.cloud_io import load as pl_load
 from torch import nn
 from torch.optim import Adam
 
-from mayavoz.data.dataset import MayaDataset
+from enhancer.data.dataset import EnhancerDataset
-from mayavoz.inference import Inference
+from enhancer.inference import Inference
-from mayavoz.loss import LOSS_MAP, LossWrapper
+from enhancer.loss import LOSS_MAP, LossWrapper
-from mayavoz.version import __version__
+from enhancer.version import __version__
 
-CACHE_DIR = os.getenv(
+CACHE_DIR = ""
-    "ENHANCER_CACHE",
+HF_TORCH_WEIGHTS = ""
-    os.path.expanduser("~/.cache/torch/mayavoz"),
-)
-HF_TORCH_WEIGHTS = "pytorch_model.ckpt"
 DEFAULT_DEVICE = "cpu"
-SAVE_NAME = "mayavoz"
 
 
-class Mayamodel(pl.LightningModule):
+class Model(pl.LightningModule):
     """
     Base class for all models
     parameters:
@@ -37,8 +33,8 @@ class Mayamodel(pl.LightningModule):
             audio sampling rate
         lr: float, optional
             learning rate for model training
-        dataset: MayaDataset, optional
+        dataset: EnhancerDataset, optional
-            mayavoz dataset used for training/validation
+            Enhancer dataset used for training/validation
         duration: float, optional
             duration used for training/inference
         loss : string or List of strings or custom loss (nn.Module), default to "mse"
@@ -51,13 +47,15 @@ class Mayamodel(pl.LightningModule):
         num_channels: int = 1,
         sampling_rate: int = 16000,
         lr: float = 1e-3,
-        dataset: Optional[MayaDataset] = None,
+        dataset: Optional[EnhancerDataset] = None,
         duration: Optional[float] = None,
         loss: Union[str, List] = "mse",
         metric: Union[str, List, Any] = "mse",
     ):
         super().__init__()
-        assert num_channels == 1, "mayavoz only support for mono channel models"
+        assert (
+            num_channels == 1
+        ), "Enhancer only support for mono channel models"
         self.dataset = dataset
         self.save_hyperparameters(
             "num_channels", "sampling_rate", "lr", "loss", "metric", "duration"
@@ -234,8 +232,8 @@ class Mayamodel(pl.LightningModule):
 
     def on_save_checkpoint(self, checkpoint):
 
-        checkpoint[SAVE_NAME] = {
+        checkpoint["enhancer"] = {
-            "version": {SAVE_NAME: __version__, "pytorch": torch.__version__},
+            "version": {"enhancer": __version__, "pytorch": torch.__version__},
             "architecture": {
                 "module": self.__class__.__module__,
                 "class": self.__class__.__name__,
@@ -288,8 +286,8 @@ class Mayamodel(pl.LightningModule):
 
         Returns
         -------
-        model : Mayamodel
+        model : Model
-            Mayamodel
+            Model
 
         See also
         --------
@@ -318,7 +316,7 @@ class Mayamodel(pl.LightningModule):
             )
             model_path_pl = cached_download(
                 url=url,
-                library_name="mayavoz",
+                library_name="enhancer",
                 library_version=__version__,
                 cache_dir=cached_dir,
                 use_auth_token=use_auth_token,
@@ -328,8 +326,8 @@ class Mayamodel(pl.LightningModule):
             map_location = torch.device(DEFAULT_DEVICE)
 
         loaded_checkpoint = pl_load(model_path_pl, map_location)
-        module_name = loaded_checkpoint[SAVE_NAME]["architecture"]["module"]
+        module_name = loaded_checkpoint["enhancer"]["architecture"]["module"]
-        class_name = loaded_checkpoint[SAVE_NAME]["architecture"]["class"]
+        class_name = loaded_checkpoint["enhancer"]["architecture"]["class"]
         module = import_module(module_name)
         Klass = getattr(module, class_name)
 

enhancer/models/waveunet.py

@@ -1,12 +1,12 @@
-import warnings
+import logging
 from typing import List, Optional, Union
 
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 
-from mayavoz.data.dataset import MayaDataset
+from enhancer.data.dataset import EnhancerDataset
-from mayavoz.models.model import Mayamodel
+from enhancer.models.model import Model
 
 
 class WavenetDecoder(nn.Module):
@@ -66,7 +66,7 @@ class WavenetEncoder(nn.Module):
         return self.encoder(waveform)
 
 
-class WaveUnet(Mayamodel):
+class WaveUnet(Model):
     """
     Wave-U-Net model from  https://arxiv.org/pdf/1811.11307.pdf
     parameters:
@@ -80,8 +80,8 @@ class WaveUnet(Mayamodel):
             sampling rate of input audio
         lr : float, defaults to 1e-3
             learning rate used for training
-        dataset: MayaDataset, optional
+        dataset: EnhancerDataset, optional
-            MayaDataset object containing train/validation data for training
+            EnhancerDataset object containing train/validation data for training
         duration : float, optional
             chunk duration in seconds
         loss : string or List of strings
@@ -97,17 +97,17 @@ class WaveUnet(Mayamodel):
         initial_output_channels: int = 24,
         sampling_rate: int = 16000,
         lr: float = 1e-3,
-        dataset: Optional[MayaDataset] = None,
+        dataset: Optional[EnhancerDataset] = None,
         duration: Optional[float] = None,
         loss: Union[str, List] = "mse",
         metric: Union[str, List] = "mse",
     ):
         duration = (
-            dataset.duration if isinstance(dataset, MayaDataset) else duration
+            dataset.duration if isinstance(dataset, EnhancerDataset) else None
         )
         if dataset is not None:
             if sampling_rate != dataset.sampling_rate:
-                warnings.warn(
+                logging.warning(
                     f"model sampling rate {sampling_rate} should match dataset sampling rate {dataset.sampling_rate}"
                 )
                 sampling_rate = dataset.sampling_rate

enhancer/utils/__init__.py

@@ -0,0 +1,3 @@
+from enhancer.utils.config import Files
+from enhancer.utils.io import Audio
+from enhancer.utils.utils import check_files

enhancer/utils/utils.py

@@ -1,7 +1,7 @@
 import os
 from typing import Optional
 
-from mayavoz.utils.config import Files
+from enhancer.utils.config import Files
 
 
 def check_files(root_dir: str, files: Files):

environment.yml

@@ -1,4 +1,4 @@
-name: mayavoz
+name: enhancer
 
 dependencies:
   - pip=21.0.1

hpc_entrypoint.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+set -e
+
+echo '----------------------------------------------------'
+echo ' SLURM_CLUSTER_NAME = '$SLURM_CLUSTER_NAME
+echo '    SLURMD_NODENAME = '$SLURMD_NODENAME
+echo '        SLURM_JOBID = '$SLURM_JOBID
+echo '     SLURM_JOB_USER = '$SLURM_JOB_USER
+echo '    SLURM_PARTITION = '$SLURM_JOB_PARTITION
+echo '  SLURM_JOB_ACCOUNT = '$SLURM_JOB_ACCOUNT
+echo '----------------------------------------------------'
+
+#TeamCity Output
+cat << EOF
+##teamcity[buildNumber '$SLURM_JOBID']
+EOF
+
+echo "Load HPC modules"
+module load anaconda
+
+echo "Activate Environment"
+source activate enhancer
+export TRANSFORMERS_OFFLINE=True
+export PYTHONPATH=${PYTHONPATH}:/scratch/c.sistc3/enhancer
+export HYDRA_FULL_ERROR=1
+
+echo $PYTHONPATH
+
+source ~/mlflow_settings.sh
+
+echo "Making temp dir"
+mkdir temp
+pwd
+
+#python transcriber/tasks/embeddings/timit.py --directory /scratch/$USER/TIMIT/data/lisa/data/timit/raw/TIMIT/TRAIN --output ./data/train
+#python transcriber/tasks/embeddings/timit.py --directory /scratch/$USER/TIMIT/data/lisa/data/timit/raw/TIMIT/TEST --output ./data/test
+# mv /scratch/c.sistc3/MS-SNSD/DNS20/CleanSpeech_testing /scratch/c.sistc3/MS-SNSD/DNS30/CleanSpeech_testing
+# mv /scratch/c.sistc3/MS-SNSD/DNS20/NoisySpeech_testing /scratch/c.sistc3/MS-SNSD/DNS30/NoisySpeech_testing
+
+echo "Start Training..."
+python enhancer/cli/train.py

mayavoz/cli/train.py

@@ -1,120 +0,0 @@
-import os
-from types import MethodType
-
-import hydra
-from hydra.utils import instantiate
-from omegaconf import DictConfig, OmegaConf
-from pytorch_lightning.callbacks import (
-    EarlyStopping,
-    LearningRateMonitor,
-    ModelCheckpoint,
-)
-from pytorch_lightning.loggers import MLFlowLogger
-from torch.optim.lr_scheduler import ReduceLROnPlateau
-
-# from torch_audiomentations import Compose, Shift
-
-os.environ["HYDRA_FULL_ERROR"] = "1"
-JOB_ID = os.environ.get("SLURM_JOBID", "0")
-
-
-@hydra.main(config_path="train_config", config_name="config")
-def train(config: DictConfig):
-
-    OmegaConf.save(config, "config.yaml")
-
-    callbacks = []
-    logger = MLFlowLogger(
-        experiment_name=config.mlflow.experiment_name,
-        run_name=config.mlflow.run_name,
-        tags={"JOB_ID": JOB_ID},
-    )
-
-    parameters = config.hyperparameters
-    # apply_augmentations = Compose(
-    #     [
-    #         Shift(min_shift=0.5, max_shift=1.0, shift_unit="seconds", p=0.5),
-    #     ]
-    # )
-
-    dataset = instantiate(config.dataset, augmentations=None)
-    model = instantiate(
-        config.model,
-        dataset=dataset,
-        lr=parameters.get("lr"),
-        loss=parameters.get("loss"),
-        metric=parameters.get("metric"),
-    )
-
-    direction = model.valid_monitor
-    checkpoint = ModelCheckpoint(
-        dirpath="./model",
-        filename=f"model_{JOB_ID}",
-        monitor="valid_loss",
-        verbose=False,
-        mode=direction,
-        every_n_epochs=1,
-    )
-    callbacks.append(checkpoint)
-    callbacks.append(LearningRateMonitor(logging_interval="epoch"))
-
-    if parameters.get("Early_stop", False):
-        early_stopping = EarlyStopping(
-            monitor="val_loss",
-            mode=direction,
-            min_delta=0.0,
-            patience=parameters.get("EarlyStopping_patience", 10),
-            strict=True,
-            verbose=False,
-        )
-        callbacks.append(early_stopping)
-
-    def configure_optimizers(self):
-        optimizer = instantiate(
-            config.optimizer,
-            lr=parameters.get("lr"),
-            params=self.parameters(),
-        )
-        scheduler = ReduceLROnPlateau(
-            optimizer=optimizer,
-            mode=direction,
-            factor=parameters.get("ReduceLr_factor", 0.1),
-            verbose=True,
-            min_lr=parameters.get("min_lr", 1e-6),
-            patience=parameters.get("ReduceLr_patience", 3),
-        )
-        return {
-            "optimizer": optimizer,
-            "lr_scheduler": scheduler,
-            "monitor": f'valid_{parameters.get("ReduceLr_monitor", "loss")}',
-        }
-
-    model.configure_optimizers = MethodType(configure_optimizers, model)
-
-    trainer = instantiate(config.trainer, logger=logger, callbacks=callbacks)
-    trainer.fit(model)
-    trainer.test(model)
-
-    logger.experiment.log_artifact(
-        logger.run_id, f"{trainer.default_root_dir}/config.yaml"
-    )
-
-    saved_location = os.path.join(
-        trainer.default_root_dir, "model", f"model_{JOB_ID}.ckpt"
-    )
-    if os.path.isfile(saved_location):
-        logger.experiment.log_artifact(logger.run_id, saved_location)
-        logger.experiment.log_param(
-            logger.run_id,
-            "num_train_steps_per_epoch",
-            dataset.train__len__() / dataset.batch_size,
-        )
-        logger.experiment.log_param(
-            logger.run_id,
-            "num_valid_steps_per_epoch",
-            dataset.val__len__() / dataset.batch_size,
-        )
-
-
-if __name__ == "__main__":
-    train()

mayavoz/cli/train_config/dataset/MS-SNSD.yaml

@@ -1,12 +0,0 @@
-_target_: mayavoz.data.dataset.MayaDataset
-name : MS-SDSD
-root_dir : /Users/shahules/Myprojects/MS-SNSD
-duration : 2.0
-sampling_rate: 16000
-batch_size: 32
-min_valid_minutes: 15
-files:
-  train_clean : CleanSpeech_training
-  test_clean : CleanSpeech_training
-  train_noisy : NoisySpeech_training
-  test_noisy : NoisySpeech_training

mayavoz/cli/train_config/dataset/Valentini.yaml

@@ -1,13 +0,0 @@
-_target_: mayavoz.data.dataset.MayaDataset
-name : Valentini
-root_dir : /scratch/c.sistc3/DS_10283_2791
-duration : 4.5
-stride : 2
-sampling_rate: 16000
-batch_size: 32
-valid_minutes : 15
-files:
-  train_clean : clean_trainset_28spk_wav
-  test_clean : clean_testset_wav
-  train_noisy : noisy_trainset_28spk_wav
-  test_noisy : noisy_testset_wav

mayavoz/cli/train_config/hyperparameters/default.yaml

@@ -1,7 +0,0 @@
-loss : mae
-metric : [stoi,pesq,si-sdr]
-lr : 0.0003
-ReduceLr_patience : 5
-ReduceLr_factor : 0.2
-min_lr : 0.000001
-EarlyStopping_factor : 10

mayavoz/cli/train_config/mlflow/experiment.yaml

@@ -1,2 +0,0 @@
-experiment_name : shahules/mayavoz
-run_name : Demucs + Vtck with stride + augmentations

mayavoz/cli/train_config/model/Demucs.yaml

@@ -1,16 +0,0 @@
-_target_: mayavoz.models.demucs.Demucs
-num_channels: 1
-resample: 4
-sampling_rate : 16000
-
-encoder_decoder:
-  depth: 4
-  initial_output_channels: 64
-  kernel_size: 8
-  stride: 4
-  growth_factor: 2
-  glu: True
-
-lstm:
-  bidirectional: False
-  num_layers: 2

mayavoz/cli/train_config/trainer/default.yaml

@@ -1,46 +0,0 @@
-_target_: pytorch_lightning.Trainer
-accelerator: gpu
-accumulate_grad_batches: 1
-amp_backend: native
-auto_lr_find: True
-auto_scale_batch_size: False
-auto_select_gpus: True
-benchmark: False
-check_val_every_n_epoch: 1
-detect_anomaly: False
-deterministic: False
-devices: 2
-enable_checkpointing: True
-enable_model_summary: True
-enable_progress_bar: True
-fast_dev_run: False
-gpus: null
-gradient_clip_val: 0
-gradient_clip_algorithm: norm
-ipus: null
-limit_predict_batches: 1.0
-limit_test_batches: 1.0
-limit_train_batches: 1.0
-limit_val_batches: 1.0
-log_every_n_steps: 50
-max_epochs: 200
-max_steps: -1
-max_time: null
-min_epochs: 1
-min_steps: null
-move_metrics_to_cpu: False
-multiple_trainloader_mode: max_size_cycle
-num_nodes: 1
-num_processes: 1
-num_sanity_val_steps: 2
-overfit_batches: 0.0
-precision: 32
-profiler: null
-reload_dataloaders_every_n_epochs: 0
-replace_sampler_ddp: True
-strategy: ddp
-sync_batchnorm: False
-tpu_cores: null
-track_grad_norm: -1
-val_check_interval: 1.0
-weights_save_path: null

mayavoz/data/__init__.py

@@ -1 +0,0 @@
-from mayavoz.data.dataset import MayaDataset

mayavoz/models/__init__.py

@@ -1,3 +0,0 @@
-from mayavoz.models.demucs import Demucs
-from mayavoz.models.model import Mayamodel
-from mayavoz.models.waveunet import WaveUnet

mayavoz/models/complexnn/__init__.py

@@ -1,5 +0,0 @@
-from mayavoz.models.complexnn.conv import ComplexConv2d  # noqa
-from mayavoz.models.complexnn.conv import ComplexConvTranspose2d  # noqa
-from mayavoz.models.complexnn.rnn import ComplexLSTM  # noqa
-from mayavoz.models.complexnn.utils import ComplexBatchNorm2D  # noqa
-from mayavoz.models.complexnn.utils import ComplexRelu  # noqa

mayavoz/utils/__init__.py

@@ -1,3 +0,0 @@
-from mayavoz.utils.config import Files
-from mayavoz.utils.io import Audio
-from mayavoz.utils.utils import check_files

notebooks/Custom_model_training.ipynb

@@ -1,338 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "markdown",
-   "id": "ccd61d5c",
-   "metadata": {},
-   "source": [
-    "## Custom model training using mayavoz [advanced]\n",
-    "\n",
-    "In this tutorial, we will cover advanced usages and customizations for training your own speecg enhancement model. \n",
-    "\n",
-    " - [Data preparation using MayaDataset](#dataprep)\n",
-    " - [Model customization](#modelcustom)\n",
-    " - [callbacks & LR schedulers](#callbacks)\n",
-    " - [Model training & testing](#train)\n"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "726c320f",
-   "metadata": {},
-   "source": [
-    "- **install mayavoz**"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "c987c799",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "! pip install -q mayavoz"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "8ff9857b",
-   "metadata": {},
-   "source": [
-    "<div id=\"dataprep\"></div>\n",
-    "\n",
-    "### Data preparation\n",
-    "\n",
-    "`Files` is a dataclass that wraps and holds train/test paths togethor. There are usually one folder each for clean and noisy data. These paths must be relative to a `root_dir` where all these directories reside. For example\n",
-    "\n",
-    "```\n",
-    "- VCTK/\n",
-    "    |__ clean_train_wav/\n",
-    "    |__ noisy_train_wav/\n",
-    "    |__ clean_test_wav/\n",
-    "    |__ noisy_test_wav/\n",
-    "    \n",
-    "```"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 2,
-   "id": "64cbc0c8",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from mayavoz.utils import Files\n",
-    "file = Files(train_clean=\"clean_train_wav\",\n",
-    "            train_noisy=\"noisy_train_wav\",\n",
-    "            test_clean=\"clean_test_wav\",\n",
-    "            test_noisy=\"noisy_test_wav\")\n",
-    "root_dir = \"VCTK\""
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "2d324bd1",
-   "metadata": {},
-   "source": [
-    "- `name`: name of the dataset. \n",
-    "- `duration`: control the duration of each audio instance fed into your model.\n",
-    "- `stride` is used if set to move the sliding window.\n",
-    "- `sampling_rate`: desired sampling rate for audio\n",
-    "- `batch_size`: model batch size\n",
-    "- `min_valid_minutes`: minimum validation in minutes. Validation is automatically selected from training set. (exclusive users).\n",
-    "- `matching_function`: there are two types of mapping functions.\n",
-    "    - `one_to_one` : In this one clean file will only have one corresponding noisy file. For example Valentini datasets\n",
-    "    - `one_to_many` : In this one clean file will only have one corresponding noisy file. For example MS-SNSD dataset.\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 3,
-   "id": "6834941d",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "name = \"vctk\"\n",
-    "duration : 4.5\n",
-    "stride : 2.0\n",
-    "sampling_rate : 16000\n",
-    "min_valid_minutes : 20.0\n",
-    "batch_size : 32\n",
-    "matching_function : \"one_to_one\"\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "d08c6bf8",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from mayavoz.dataset import MayaDataset\n",
-    "dataset = MayaDataset(\n",
-    "            name=name,\n",
-    "            root_dir=root_dir,\n",
-    "            files=files,\n",
-    "            duration=duration,\n",
-    "            stride=stride,\n",
-    "            sampling_rate=sampling_rate,\n",
-    "            batch_size=batch_size,\n",
-    "            min_valid_minutes=min_valid_minutes,\n",
-    "            matching_function=matching_function\n",
-    "        )"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "5b315bde",
-   "metadata": {},
-   "source": [
-    "Now your custom dataloader is ready!"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "01548fe5",
-   "metadata": {},
-   "source": [
-    "<div id=\"modelcustom\"></div>\n",
-    "\n",
-    "### Model Customization\n",
-    "Now, this is very easy. \n",
-    "\n",
-    "- Import the preferred model from `mayavoz.models`. Currently 3 models are implemented.\n",
-    "   - `WaveUnet`\n",
-    "   - `Demucs`\n",
-    "   - `DCCRN`\n",
-    "- Each of model hyperparameters such as depth,kernel_size,stride etc can be controlled by you.   Just check the parameters and pass it to as required.\n",
-    "- `sampling_rate`: sampling rate (should be equal to dataset sampling rate)\n",
-    "- `dataset`: mayavoz dataset object as prepared earlier.\n",
-    "- `loss` : model loss. Multiple loss functions are available.\n",
-    "\n",
-    "        \n",
-    "        \n",
-    "you can pass one (as string)/more (as list of strings) of these loss functions as per your requirements. For example, model will automatically calculate loss as average of `mae` and `mse` if you pass loss as `[\"mae\",\"mse\"]`. Available loss functions are `mse`,`mae`,`si-snr`.\n",
-    "\n",
-    "mayavoz can accept **custom loss functions**. It should be of the form.\n",
-    "```\n",
-    "class your_custom_loss(nn.Module):\n",
-    "    def __init__(self,**kwargs):\n",
-    "        self.higher_better = False  ## loss minimization direction\n",
-    "        self.name = \"your_loss_name\" ## loss name logging \n",
-    "        ...\n",
-    "    def forward(self,prediction, target):\n",
-    "        loss = ....\n",
-    "        return loss\n",
-    "        \n",
-    "```\n",
-    "\n",
-    "- metrics : validation metrics. Available options `mae`,`mse`,`si-sdr`,`si-sdr`,`pesq`,`stoi`. One or more can be used.\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "b36b457c",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from mayavoz.models import Demucs\n",
-    "model = Demucs(\n",
-    "        sampling_rate=16000,\n",
-    "        dataset=dataset,\n",
-    "        loss=[\"mae\"],\n",
-    "        metrics=[\"stoi\",\"pesq\"])\n"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "1523d638",
-   "metadata": {},
-   "source": [
-    "<div id=\"callbacks\"></div>\n",
-    "\n",
-    "### learning rate schedulers and callbacks\n",
-    "Here I am using `ReduceLROnPlateau`"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "8de6931c",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from torch.optim.lr_scheduler import ReduceLROnPlateau\n",
-    "\n",
-    "def configure_optimizers(self):\n",
-    "        optimizer = instantiate(\n",
-    "            config.optimizer,\n",
-    "            lr=parameters.get(\"lr\"),\n",
-    "            params=self.parameters(),\n",
-    "        )\n",
-    "        scheduler = ReduceLROnPlateau(\n",
-    "            optimizer=optimizer,\n",
-    "            mode=direction,\n",
-    "            factor=parameters.get(\"ReduceLr_factor\", 0.1),\n",
-    "            verbose=True,\n",
-    "            min_lr=parameters.get(\"min_lr\", 1e-6),\n",
-    "            patience=parameters.get(\"ReduceLr_patience\", 3),\n",
-    "        )\n",
-    "        return {\n",
-    "            \"optimizer\": optimizer,\n",
-    "            \"lr_scheduler\": scheduler,\n",
-    "            \"monitor\": f'valid_{parameters.get(\"ReduceLr_monitor\", \"loss\")}',\n",
-    "        }\n",
-    "\n",
-    "\n",
-    "model.configure_optimizers = MethodType(configure_optimizers, model)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "2f7b5af5",
-   "metadata": {},
-   "source": [
-    "you can use any number of callbacks and pass it directly to pytorch lightning trainer. Here I am using only `ModelCheckpoint`"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "6f6b62a1",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "callbacks = []\n",
-    "direction = model.valid_monitor ## min or max \n",
-    "checkpoint = ModelCheckpoint(\n",
-    "        dirpath=\"./model\",\n",
-    "        filename=f\"model_filename\",\n",
-    "        monitor=\"valid_loss\",\n",
-    "        verbose=False,\n",
-    "        mode=direction,\n",
-    "        every_n_epochs=1,\n",
-    "    )\n",
-    "callbacks.append(checkpoint)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "f3534445",
-   "metadata": {},
-   "source": [
-    "<div id=\"train\"></div>\n",
-    "\n",
-    "\n",
-    "### Train"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "3dc0348b",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import pytorch_lightning as pl\n",
-    "trainer = plt.Trainer(max_epochs=1,callbacks=callbacks,accelerator=\"gpu\")\n",
-    "trainer.fit(model)\n"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "56dcfec1",
-   "metadata": {},
-   "source": [
-    "- Test your model agaist test dataset"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "63851feb",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "trainer.test(model)"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "4d3f5350",
-   "metadata": {},
-   "source": [
-    "**Hurray! you have your speech enhancement model trained and tested.**\n"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "10d630e8",
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  }
- ],
- "metadata": {
-  "kernelspec": {
-   "display_name": "enhancer",
-   "language": "python",
-   "name": "enhancer"
-  },
-  "language_info": {
-   "codemirror_mode": {
-    "name": "ipython",
-    "version": 3
-   },
-   "file_extension": ".py",
-   "mimetype": "text/x-python",
-   "name": "python",
-   "nbconvert_exporter": "python",
-   "pygments_lexer": "ipython3",
-   "version": "3.8.13"
-  }
- },
- "nbformat": 4,
- "nbformat_minor": 5
-}

recipes/MS-SNSD/MS-SNSD-20/DCCRN/train.py

@@ -1,120 +0,0 @@
-import os
-from types import MethodType
-
-import hydra
-from hydra.utils import instantiate
-from omegaconf import DictConfig, OmegaConf
-from pytorch_lightning.callbacks import (
-    EarlyStopping,
-    LearningRateMonitor,
-    ModelCheckpoint,
-)
-from pytorch_lightning.loggers import MLFlowLogger
-from torch.optim.lr_scheduler import ReduceLROnPlateau
-
-# from torch_audiomentations import Compose, Shift
-
-os.environ["HYDRA_FULL_ERROR"] = "1"
-JOB_ID = os.environ.get("SLURM_JOBID", "0")
-
-
-@hydra.main(config_path="train_config", config_name="config")
-def train(config: DictConfig):
-
-    OmegaConf.save(config, "config.yaml")
-
-    callbacks = []
-    logger = MLFlowLogger(
-        experiment_name=config.mlflow.experiment_name,
-        run_name=config.mlflow.run_name,
-        tags={"JOB_ID": JOB_ID},
-    )
-
-    parameters = config.hyperparameters
-    # apply_augmentations = Compose(
-    #     [
-    #         Shift(min_shift=0.5, max_shift=1.0, shift_unit="seconds", p=0.5),
-    #     ]
-    # )
-
-    dataset = instantiate(config.dataset, augmentations=None)
-    model = instantiate(
-        config.model,
-        dataset=dataset,
-        lr=parameters.get("lr"),
-        loss=parameters.get("loss"),
-        metric=parameters.get("metric"),
-    )
-
-    direction = model.valid_monitor
-    checkpoint = ModelCheckpoint(
-        dirpath="./model",
-        filename=f"model_{JOB_ID}",
-        monitor="valid_loss",
-        verbose=False,
-        mode=direction,
-        every_n_epochs=1,
-    )
-    callbacks.append(checkpoint)
-    callbacks.append(LearningRateMonitor(logging_interval="epoch"))
-
-    if parameters.get("Early_stop", False):
-        early_stopping = EarlyStopping(
-            monitor="val_loss",
-            mode=direction,
-            min_delta=0.0,
-            patience=parameters.get("EarlyStopping_patience", 10),
-            strict=True,
-            verbose=False,
-        )
-        callbacks.append(early_stopping)
-
-    def configure_optimizers(self):
-        optimizer = instantiate(
-            config.optimizer,
-            lr=parameters.get("lr"),
-            params=self.parameters(),
-        )
-        scheduler = ReduceLROnPlateau(
-            optimizer=optimizer,
-            mode=direction,
-            factor=parameters.get("ReduceLr_factor", 0.1),
-            verbose=True,
-            min_lr=parameters.get("min_lr", 1e-6),
-            patience=parameters.get("ReduceLr_patience", 3),
-        )
-        return {
-            "optimizer": optimizer,
-            "lr_scheduler": scheduler,
-            "monitor": f'valid_{parameters.get("ReduceLr_monitor", "loss")}',
-        }
-
-    model.configure_optimizers = MethodType(configure_optimizers, model)
-
-    trainer = instantiate(config.trainer, logger=logger, callbacks=callbacks)
-    trainer.fit(model)
-    trainer.test(model)
-
-    logger.experiment.log_artifact(
-        logger.run_id, f"{trainer.default_root_dir}/config.yaml"
-    )
-
-    saved_location = os.path.join(
-        trainer.default_root_dir, "model", f"model_{JOB_ID}.ckpt"
-    )
-    if os.path.isfile(saved_location):
-        logger.experiment.log_artifact(logger.run_id, saved_location)
-        logger.experiment.log_param(
-            logger.run_id,
-            "num_train_steps_per_epoch",
-            dataset.train__len__() / dataset.batch_size,
-        )
-        logger.experiment.log_param(
-            logger.run_id,
-            "num_valid_steps_per_epoch",
-            dataset.val__len__() / dataset.batch_size,
-        )
-
-
-if __name__ == "__main__":
-    train()

recipes/MS-SNSD/MS-SNSD-20/DCCRN/train_config/config.yaml

@@ -1,7 +0,0 @@
-defaults:
-  - model : Demucs
-  - dataset : MS-SNSD
-  - optimizer : Adam
-  - hyperparameters : default
-  - trainer : default
-  - mlflow : experiment

recipes/MS-SNSD/MS-SNSD-20/DCCRN/train_config/dataset/MS-SNSD.yaml

@@ -1,13 +0,0 @@
-_target_: mayavoz.data.dataset.MayaDataset
-name : MS-SDSD
-root_dir : /Users/shahules/Myprojects/MS-SNSD
-duration : 1.5
-stride : 1
-sampling_rate: 16000
-batch_size: 32
-min_valid_minutes: 25
-files:
-  train_clean : CleanSpeech_training
-  test_clean : CleanSpeech_training
-  train_noisy : NoisySpeech_training
-  test_noisy : NoisySpeech_training

recipes/MS-SNSD/MS-SNSD-20/DCCRN/train_config/mlflow/experiment.yaml

@@ -1,2 +0,0 @@
-experiment_name : shahules/mayavoz
-run_name : Demucs + Vtck with stride + augmentations

recipes/MS-SNSD/MS-SNSD-20/DCCRN/train_config/model/DCCRN.yaml

@@ -1,25 +0,0 @@
-_target_: mayavoz.models.dccrn.DCCRN
-num_channels: 1
-sampling_rate : 16000
-complex_lstm : True
-complex_norm : True
-complex_relu : True
-masking_mode : True
-
-encoder_decoder:
-  initial_output_channels : 32
-  depth : 6
-  kernel_size : 5
-  growth_factor : 2
-  stride : 2
-  padding : 2
-  output_padding : 1
-
-lstm:
-  num_layers : 2
-  hidden_size : 256
-
-stft:
-  window_len : 400
-  hop_size : 100
-  nfft : 512

recipes/MS-SNSD/MS-SNSD-20/DCCRN/train_config/optimizer/Adam.yaml

@@ -1,6 +0,0 @@
-_target_: torch.optim.Adam
-lr: 1e-3
-betas: [0.9, 0.999]
-eps: 1e-08
-weight_decay: 0
-amsgrad: False

recipes/MS-SNSD/MS-SNSD-20/DCCRN/train_config/trainer/default.yaml

@@ -1,46 +0,0 @@
-_target_: pytorch_lightning.Trainer
-accelerator: gpu
-accumulate_grad_batches: 1
-amp_backend: native
-auto_lr_find: True
-auto_scale_batch_size: False
-auto_select_gpus: True
-benchmark: False
-check_val_every_n_epoch: 1
-detect_anomaly: False
-deterministic: False
-devices: 2
-enable_checkpointing: True
-enable_model_summary: True
-enable_progress_bar: True
-fast_dev_run: False
-gpus: null
-gradient_clip_val: 0
-gradient_clip_algorithm: norm
-ipus: null
-limit_predict_batches: 1.0
-limit_test_batches: 1.0
-limit_train_batches: 1.0
-limit_val_batches: 1.0
-log_every_n_steps: 50
-max_epochs: 200
-max_steps: -1
-max_time: null
-min_epochs: 1
-min_steps: null
-move_metrics_to_cpu: False
-multiple_trainloader_mode: max_size_cycle
-num_nodes: 1
-num_processes: 1
-num_sanity_val_steps: 2
-overfit_batches: 0.0
-precision: 32
-profiler: null
-reload_dataloaders_every_n_epochs: 0
-replace_sampler_ddp: True
-strategy: ddp
-sync_batchnorm: False
-tpu_cores: null
-track_grad_norm: -1
-val_check_interval: 1.0
-weights_save_path: null

recipes/MS-SNSD/MS-SNSD-20/DCCRN/train_config/trainer/fastrun_dev.yaml

@@ -1,2 +0,0 @@
-_target_: pytorch_lightning.Trainer
-fast_dev_run: True

recipes/MS-SNSD/MS-SNSD-20/Demucs/train.py

@@ -1,120 +0,0 @@
-import os
-from types import MethodType
-
-import hydra
-from hydra.utils import instantiate
-from omegaconf import DictConfig, OmegaConf
-from pytorch_lightning.callbacks import (
-    EarlyStopping,
-    LearningRateMonitor,
-    ModelCheckpoint,
-)
-from pytorch_lightning.loggers import MLFlowLogger
-from torch.optim.lr_scheduler import ReduceLROnPlateau
-
-# from torch_audiomentations import Compose, Shift
-
-os.environ["HYDRA_FULL_ERROR"] = "1"
-JOB_ID = os.environ.get("SLURM_JOBID", "0")
-
-
-@hydra.main(config_path="train_config", config_name="config")
-def train(config: DictConfig):
-
-    OmegaConf.save(config, "config.yaml")
-
-    callbacks = []
-    logger = MLFlowLogger(
-        experiment_name=config.mlflow.experiment_name,
-        run_name=config.mlflow.run_name,
-        tags={"JOB_ID": JOB_ID},
-    )
-
-    parameters = config.hyperparameters
-    # apply_augmentations = Compose(
-    #     [
-    #         Shift(min_shift=0.5, max_shift=1.0, shift_unit="seconds", p=0.5),
-    #     ]
-    # )
-
-    dataset = instantiate(config.dataset, augmentations=None)
-    model = instantiate(
-        config.model,
-        dataset=dataset,
-        lr=parameters.get("lr"),
-        loss=parameters.get("loss"),
-        metric=parameters.get("metric"),
-    )
-
-    direction = model.valid_monitor
-    checkpoint = ModelCheckpoint(
-        dirpath="./model",
-        filename=f"model_{JOB_ID}",
-        monitor="valid_loss",
-        verbose=False,
-        mode=direction,
-        every_n_epochs=1,
-    )
-    callbacks.append(checkpoint)
-    callbacks.append(LearningRateMonitor(logging_interval="epoch"))
-
-    if parameters.get("Early_stop", False):
-        early_stopping = EarlyStopping(
-            monitor="val_loss",
-            mode=direction,
-            min_delta=0.0,
-            patience=parameters.get("EarlyStopping_patience", 10),
-            strict=True,
-            verbose=False,
-        )
-        callbacks.append(early_stopping)
-
-    def configure_optimizers(self):
-        optimizer = instantiate(
-            config.optimizer,
-            lr=parameters.get("lr"),
-            params=self.parameters(),
-        )
-        scheduler = ReduceLROnPlateau(
-            optimizer=optimizer,
-            mode=direction,
-            factor=parameters.get("ReduceLr_factor", 0.1),
-            verbose=True,
-            min_lr=parameters.get("min_lr", 1e-6),
-            patience=parameters.get("ReduceLr_patience", 3),
-        )
-        return {
-            "optimizer": optimizer,
-            "lr_scheduler": scheduler,
-            "monitor": f'valid_{parameters.get("ReduceLr_monitor", "loss")}',
-        }
-
-    model.configure_optimizers = MethodType(configure_optimizers, model)
-
-    trainer = instantiate(config.trainer, logger=logger, callbacks=callbacks)
-    trainer.fit(model)
-    trainer.test(model)
-
-    logger.experiment.log_artifact(
-        logger.run_id, f"{trainer.default_root_dir}/config.yaml"
-    )
-
-    saved_location = os.path.join(
-        trainer.default_root_dir, "model", f"model_{JOB_ID}.ckpt"
-    )
-    if os.path.isfile(saved_location):
-        logger.experiment.log_artifact(logger.run_id, saved_location)
-        logger.experiment.log_param(
-            logger.run_id,
-            "num_train_steps_per_epoch",
-            dataset.train__len__() / dataset.batch_size,
-        )
-        logger.experiment.log_param(
-            logger.run_id,
-            "num_valid_steps_per_epoch",
-            dataset.val__len__() / dataset.batch_size,
-        )
-
-
-if __name__ == "__main__":
-    train()

recipes/MS-SNSD/MS-SNSD-20/Demucs/train_config/config.yaml

@@ -1,7 +0,0 @@
-defaults:
-  - model : Demucs
-  - dataset : MS-SNSD
-  - optimizer : Adam
-  - hyperparameters : default
-  - trainer : default
-  - mlflow : experiment

recipes/MS-SNSD/MS-SNSD-20/Demucs/train_config/dataset/MS-SNSD.yaml

@@ -1,13 +0,0 @@
-_target_: mayavoz.data.dataset.MayaDataset
-name : MS-SDSD
-root_dir : /Users/shahules/Myprojects/MS-SNSD
-duration : 5
-stride : 1
-sampling_rate: 16000
-batch_size: 32
-min_valid_minutes: 25
-files:
-  train_clean : CleanSpeech_training
-  test_clean : CleanSpeech_training
-  train_noisy : NoisySpeech_training
-  test_noisy : NoisySpeech_training

recipes/MS-SNSD/MS-SNSD-20/Demucs/train_config/hyperparameters/default.yaml

@@ -1,7 +0,0 @@
-loss : mae
-metric : [stoi,pesq]
-lr : 0.0003
-ReduceLr_patience : 10
-ReduceLr_factor : 0.5
-min_lr : 0.000001
-EarlyStopping_factor : 10

recipes/MS-SNSD/MS-SNSD-20/Demucs/train_config/mlflow/experiment.yaml

@@ -1,2 +0,0 @@
-experiment_name : shahules/mayavoz
-run_name : demucs-ms-snsd

recipes/MS-SNSD/MS-SNSD-20/Demucs/train_config/model/Demucs.yaml

@@ -1,16 +0,0 @@
-_target_: mayavoz.models.demucs.Demucs
-num_channels: 1
-resample: 4
-sampling_rate : 16000
-
-encoder_decoder:
-  depth: 4
-  initial_output_channels: 64
-  kernel_size: 8
-  stride: 4
-  growth_factor: 2
-  glu: True
-
-lstm:
-  bidirectional: False
-  num_layers: 2

recipes/MS-SNSD/MS-SNSD-20/Demucs/train_config/optimizer/Adam.yaml

@@ -1,6 +0,0 @@
-_target_: torch.optim.Adam
-lr: 1e-3
-betas: [0.9, 0.999]
-eps: 1e-08
-weight_decay: 0
-amsgrad: False

recipes/MS-SNSD/MS-SNSD-20/Demucs/train_config/trainer/default.yaml

@@ -1,46 +0,0 @@
-_target_: pytorch_lightning.Trainer
-accelerator: gpu
-accumulate_grad_batches: 1
-amp_backend: native
-auto_lr_find: True
-auto_scale_batch_size: False
-auto_select_gpus: True
-benchmark: False
-check_val_every_n_epoch: 1
-detect_anomaly: False
-deterministic: False
-devices: 2
-enable_checkpointing: True
-enable_model_summary: True
-enable_progress_bar: True
-fast_dev_run: False
-gpus: null
-gradient_clip_val: 0
-gradient_clip_algorithm: norm
-ipus: null
-limit_predict_batches: 1.0
-limit_test_batches: 1.0
-limit_train_batches: 1.0
-limit_val_batches: 1.0
-log_every_n_steps: 50
-max_epochs: 200
-max_steps: -1
-max_time: null
-min_epochs: 1
-min_steps: null
-move_metrics_to_cpu: False
-multiple_trainloader_mode: max_size_cycle
-num_nodes: 1
-num_processes: 1
-num_sanity_val_steps: 2
-overfit_batches: 0.0
-precision: 32
-profiler: null
-reload_dataloaders_every_n_epochs: 0
-replace_sampler_ddp: True
-strategy: ddp
-sync_batchnorm: False
-tpu_cores: null
-track_grad_norm: -1
-val_check_interval: 1.0
-weights_save_path: null

recipes/MS-SNSD/MS-SNSD-20/Demucs/train_config/trainer/fastrun_dev.yaml

@@ -1,2 +0,0 @@
-_target_: pytorch_lightning.Trainer
-fast_dev_run: True

recipes/MS-SNSD/MS-SNSD-20/README.md

@@ -1,17 +0,0 @@
-### Microsoft Scalable Noisy Speech Dataset (MS-SNSD)
-
- MS-SNSD is a speech datasetthat can scale to arbitrary sizes depending on the number of speakers, noise types, and Speech to Noise Ratio (SNR) levels desired.
-
-### Dataset download & setup
-- Follow steps in the official repo [here](https://github.com/microsoft/MS-SNSD) to download and setup the dataset.
-
-**References**
-```BibTex
-@article{reddy2019scalable,
-  title={A Scalable Noisy Speech Dataset and Online Subjective Test Framework},
-  author={Reddy, Chandan KA and Beyrami, Ebrahim and Pool, Jamie and Cutler, Ross and Srinivasan, Sriram and Gehrke, Johannes},
-  journal={Proc. Interspeech 2019},
-  pages={1816--1820},
-  year={2019}
-}
-```

recipes/Valentini-dataset/28spk/Demucs/train.py

@@ -1,120 +0,0 @@
-import os
-from types import MethodType
-
-import hydra
-from hydra.utils import instantiate
-from omegaconf import DictConfig, OmegaConf
-from pytorch_lightning.callbacks import (
-    EarlyStopping,
-    LearningRateMonitor,
-    ModelCheckpoint,
-)
-from pytorch_lightning.loggers import MLFlowLogger
-from torch.optim.lr_scheduler import ReduceLROnPlateau
-
-# from torch_audiomentations import Compose, Shift
-
-os.environ["HYDRA_FULL_ERROR"] = "1"
-JOB_ID = os.environ.get("SLURM_JOBID", "0")
-
-
-@hydra.main(config_path="train_config", config_name="config")
-def main(config: DictConfig):
-
-    OmegaConf.save(config, "config_log.yaml")
-
-    callbacks = []
-    logger = MLFlowLogger(
-        experiment_name=config.mlflow.experiment_name,
-        run_name=config.mlflow.run_name,
-        tags={"JOB_ID": JOB_ID},
-    )
-
-    parameters = config.hyperparameters
-    # apply_augmentations = Compose(
-    #     [
-    #         Shift(min_shift=0.5, max_shift=1.0, shift_unit="seconds", p=0.5),
-    #     ]
-    # )
-
-    dataset = instantiate(config.dataset, augmentations=None)
-    model = instantiate(
-        config.model,
-        dataset=dataset,
-        lr=parameters.get("lr"),
-        loss=parameters.get("loss"),
-        metric=parameters.get("metric"),
-    )
-
-    direction = model.valid_monitor
-    checkpoint = ModelCheckpoint(
-        dirpath="./model",
-        filename=f"model_{JOB_ID}",
-        monitor="valid_loss",
-        verbose=False,
-        mode=direction,
-        every_n_epochs=1,
-    )
-    callbacks.append(checkpoint)
-    callbacks.append(LearningRateMonitor(logging_interval="epoch"))
-
-    if parameters.get("Early_stop", False):
-        early_stopping = EarlyStopping(
-            monitor="val_loss",
-            mode=direction,
-            min_delta=0.0,
-            patience=parameters.get("EarlyStopping_patience", 10),
-            strict=True,
-            verbose=False,
-        )
-        callbacks.append(early_stopping)
-
-    def configure_optimizers(self):
-        optimizer = instantiate(
-            config.optimizer,
-            lr=parameters.get("lr"),
-            params=self.parameters(),
-        )
-        scheduler = ReduceLROnPlateau(
-            optimizer=optimizer,
-            mode=direction,
-            factor=parameters.get("ReduceLr_factor", 0.1),
-            verbose=True,
-            min_lr=parameters.get("min_lr", 1e-6),
-            patience=parameters.get("ReduceLr_patience", 3),
-        )
-        return {
-            "optimizer": optimizer,
-            "lr_scheduler": scheduler,
-            "monitor": f'valid_{parameters.get("ReduceLr_monitor", "loss")}',
-        }
-
-    model.configure_optimizers = MethodType(configure_optimizers, model)
-
-    trainer = instantiate(config.trainer, logger=logger, callbacks=callbacks)
-    trainer.fit(model)
-    trainer.test(model)
-
-    logger.experiment.log_artifact(
-        logger.run_id, f"{trainer.default_root_dir}/config_log.yaml"
-    )
-
-    saved_location = os.path.join(
-        trainer.default_root_dir, "model", f"model_{JOB_ID}.ckpt"
-    )
-    if os.path.isfile(saved_location):
-        logger.experiment.log_artifact(logger.run_id, saved_location)
-        logger.experiment.log_param(
-            logger.run_id,
-            "num_train_steps_per_epoch",
-            dataset.train__len__() / dataset.batch_size,
-        )
-        logger.experiment.log_param(
-            logger.run_id,
-            "num_valid_steps_per_epoch",
-            dataset.val__len__() / dataset.batch_size,
-        )
-
-
-if __name__ == "__main__":
-    main()

recipes/Valentini-dataset/28spk/Demucs/train_config/config.yaml

@@ -1,7 +0,0 @@
-defaults:
-  - model : Demucs
-  - dataset : Vctk
-  - optimizer : Adam
-  - hyperparameters : default
-  - trainer : default
-  - mlflow : experiment

recipes/Valentini-dataset/28spk/Demucs/train_config/hyperparameters/default.yaml

@@ -1,8 +0,0 @@
-loss : mae
-metric : [stoi,pesq,si-sdr]
-lr : 0.0003
-Early_stop : False
-ReduceLr_patience : 10
-ReduceLr_factor : 0.1
-min_lr : 0.000001
-EarlyStopping_factor : 10

recipes/Valentini-dataset/28spk/Demucs/train_config/mlflow/experiment.yaml

@@ -1,2 +0,0 @@
-experiment_name : shahules/mayavoz
-run_name : baseline

recipes/Valentini-dataset/28spk/Demucs/train_config/optimizer/Adam.yaml

@@ -1,6 +0,0 @@
-_target_: torch.optim.Adam
-lr: 1e-3
-betas: [0.9, 0.999]
-eps: 1e-08
-weight_decay: 0
-amsgrad: False

recipes/Valentini-dataset/28spk/WaveUnet/train_config/config.yaml

@@ -1,7 +0,0 @@
-defaults:
-  - model : WaveUnet
-  - dataset : Vctk
-  - optimizer : Adam
-  - hyperparameters : default
-  - trainer : default
-  - mlflow : experiment

recipes/Valentini-dataset/28spk/WaveUnet/train_config/dataset/Vctk.yaml

@@ -1,13 +0,0 @@
-_target_: mayavoz.data.dataset.MayaDataset
-name : vctk
-root_dir : /scratch/c.sistc3/DS_10283_2791
-duration : 2
-stride : 1
-sampling_rate: 16000
-batch_size: 128
-valid_minutes : 25
-files:
-  train_clean : clean_trainset_28spk_wav
-  test_clean : clean_testset_wav
-  train_noisy : noisy_trainset_28spk_wav
-  test_noisy : noisy_testset_wav

recipes/Valentini-dataset/28spk/WaveUnet/train_config/hyperparameters/default.yaml

@@ -1,8 +0,0 @@
-loss : mae
-metric : [stoi,pesq,si-sdr]
-lr : 0.003
-ReduceLr_patience : 10
-ReduceLr_factor : 0.1
-min_lr : 0.000001
-EarlyStopping_factor : 10
-Early_stop : False

recipes/Valentini-dataset/28spk/WaveUnet/train_config/mlflow/experiment.yaml

@@ -1,2 +0,0 @@
-experiment_name : shahules/mayavoz
-run_name : baseline

recipes/Valentini-dataset/28spk/WaveUnet/train_config/model/WaveUnet.yaml

@@ -1,5 +0,0 @@
-_target_: mayavoz.models.waveunet.WaveUnet
-num_channels : 1
-depth : 9
-initial_output_channels: 24
-sampling_rate : 16000

recipes/Valentini-dataset/28spk/WaveUnet/train_config/optimizer/Adam.yaml

@@ -1,6 +0,0 @@
-_target_: torch.optim.Adam
-lr: 1e-3
-betas: [0.9, 0.999]
-eps: 1e-08
-weight_decay: 0
-amsgrad: False

recipes/Valentini-dataset/28spk/WaveUnet/train_config/trainer/default.yaml

@@ -1,46 +0,0 @@
-_target_: pytorch_lightning.Trainer
-accelerator: gpu
-accumulate_grad_batches: 1
-amp_backend: native
-auto_lr_find: True
-auto_scale_batch_size: False
-auto_select_gpus: True
-benchmark: False
-check_val_every_n_epoch: 1
-detect_anomaly: False
-deterministic: False
-devices: 2
-enable_checkpointing: True
-enable_model_summary: True
-enable_progress_bar: True
-fast_dev_run: False
-gpus: null
-gradient_clip_val: 0
-gradient_clip_algorithm: norm
-ipus: null
-limit_predict_batches: 1.0
-limit_test_batches: 1.0
-limit_train_batches: 1.0
-limit_val_batches: 1.0
-log_every_n_steps: 50
-max_epochs: 200
-max_steps: -1
-max_time: null
-min_epochs: 1
-min_steps: null
-move_metrics_to_cpu: False
-multiple_trainloader_mode: max_size_cycle
-num_nodes: 1
-num_processes: 1
-num_sanity_val_steps: 2
-overfit_batches: 0.0
-precision: 32
-profiler: null
-reload_dataloaders_every_n_epochs: 0
-replace_sampler_ddp: True
-strategy: ddp
-sync_batchnorm: False
-tpu_cores: null
-track_grad_norm: -1
-val_check_interval: 1.0
-weights_save_path: null

recipes/Valentini-dataset/28spk/WaveUnet/train_config/trainer/fastrun_dev.yaml

@@ -1,2 +0,0 @@
-_target_: pytorch_lightning.Trainer
-fast_dev_run: True

recipes/Valentini-dataset/README.md

@@ -1,12 +0,0 @@
-## Valentini dataset
-
-Clean and noisy parallel speech database. The database was designed to train and test speech enhancement methods that operate at 48kHz. A more detailed description can be found in the papers associated with the database.[official page](https://datashare.ed.ac.uk/handle/10283/2791)
-
-**References**
-```BibTex
-@misc{
-title={Noisy speech database for training speech enhancement algorithms and TTS models},
-author={Valentini-Botinhao, Cassia}, year={2017},
-doi=https://doi.org/10.7488/ds/2117,
-}
-```

requirements.txt

@@ -3,7 +3,7 @@ huggingface-hub>=0.10.0
 hydra-core>=1.2.0
 joblib>=1.2.0
 librosa>=0.9.2
-mlflow>=1.28.0
+mlflow>=1.29.0
 numpy>=1.23.3
 pesq==0.0.4
 protobuf>=3.19.6

setup.cfg

@@ -3,7 +3,7 @@
 # http://setuptools.readthedocs.io/en/latest/setuptools.html#configuring-setup-using-setup-cfg-files
 
 [metadata]
-name = mayavoz
+name = enhancer
 description = Deep learning for speech enhacement
 author = Shahul Ess
 author-email = shahules786@gmail.com
@@ -53,7 +53,7 @@ cli =
 [options.entry_points]
 
 console_scripts =
-    mayavoz-train=mayavoz.cli.train:train
+    enhancer-train=enhancer.cli.train:train
 
 [test]
 # py.test options when running `python setup.py test`
@@ -66,7 +66,7 @@ extras = True
 # e.g. --cov-report html (or xml) for html/xml output or --junitxml junit.xml
 # in order to write a coverage file that can be read by Jenkins.
 addopts =
-    --cov mayavoz --cov-report term-missing
+    --cov enhancer --cov-report term-missing
     --verbose
 norecursedirs =
     dist
@@ -98,7 +98,3 @@ exclude =
     build
     dist
     .eggs
-
-[options.data_files]
-. = requirements.txt
-_ = version.txt

setup.py

@@ -33,15 +33,15 @@ elif sha != "Unknown":
     version += "+" + sha[:7]
 print("-- Building version " + version)
 
-version_path = ROOT_DIR / "mayavoz" / "version.py"
+version_path = ROOT_DIR / "enhancer" / "version.py"
 
 with open(version_path, "w") as f:
     f.write("__version__ = '{}'\n".format(version))
 
 if __name__ == "__main__":
     setup(
-        name="mayavoz",
+        name="enhancer",
-        namespace_packages=["mayavoz"],
+        namespace_packages=["enhancer"],
         version=version,
         packages=find_packages(),
         install_requires=requirements,

setup.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+set -e
+
+echo "Loading Anaconda Module"
+module load anaconda
+
+echo "Creating Virtual Environment"
+conda env create -f environment.yml ||  conda env update -f environment.yml
+
+source activate enhancer
+
+echo "copying files"
+# cp /scratch/$USER/TIMIT/.* /deep-transcriber

tests/loss_function_test.py

@@ -1,7 +1,7 @@
 import pytest
 import torch
 
-from mayavoz.loss import mean_absolute_error, mean_squared_error
+from enhancer.loss import mean_absolute_error, mean_squared_error
 
 loss_functions = [mean_absolute_error(), mean_squared_error()]
 

tests/models/complexnn_test.py

@@ -1,8 +1,8 @@
 import torch
 
-from mayavoz.models.complexnn.conv import ComplexConv2d, ComplexConvTranspose2d
+from enhancer.models.complexnn.conv import ComplexConv2d, ComplexConvTranspose2d
-from mayavoz.models.complexnn.rnn import ComplexLSTM
+from enhancer.models.complexnn.rnn import ComplexLSTM
-from mayavoz.models.complexnn.utils import ComplexBatchNorm2D
+from enhancer.models.complexnn.utils import ComplexBatchNorm2D
 
 
 def test_complexconv2d():

tests/models/demucs_test.py

@@ -1,9 +1,9 @@
 import pytest
 import torch
 
-from mayavoz.data.dataset import MayaDataset
+from enhancer.data.dataset import EnhancerDataset
-from mayavoz.models import Demucs
+from enhancer.models import Demucs
-from mayavoz.utils.config import Files
+from enhancer.utils.config import Files
 
 
 @pytest.fixture
@@ -15,9 +15,7 @@ def vctk_dataset():
         test_clean="clean_testset_wav",
         test_noisy="noisy_testset_wav",
     )
-    dataset = MayaDataset(
+    dataset = EnhancerDataset(name="vctk", root_dir=root_dir, files=files)
-        name="vctk", root_dir=root_dir, files=files, sampling_rate=16000
-    )
     return dataset
 
 

tests/models/test_dccrn.py

@@ -1,9 +1,9 @@
 import pytest
 import torch
 
-from mayavoz.data.dataset import MayaDataset
+from enhancer.data.dataset import EnhancerDataset
-from mayavoz.models.dccrn import DCCRN
+from enhancer.models.dccrn import DCCRN
-from mayavoz.utils.config import Files
+from enhancer.utils.config import Files
 
 
 @pytest.fixture
@@ -15,9 +15,7 @@ def vctk_dataset():
         test_clean="clean_testset_wav",
         test_noisy="noisy_testset_wav",
     )
-    dataset = MayaDataset(
+    dataset = EnhancerDataset(name="vctk", root_dir=root_dir, files=files)
-        name="vctk", root_dir=root_dir, files=files, sampling_rate=16000
-    )
     return dataset
 
 

tests/models/test_waveunet.py

@@ -1,9 +1,9 @@
 import pytest
 import torch
 
-from mayavoz.data.dataset import MayaDataset
+from enhancer.data.dataset import EnhancerDataset
-from mayavoz.models import WaveUnet
+from enhancer.models import WaveUnet
-from mayavoz.utils.config import Files
+from enhancer.utils.config import Files
 
 
 @pytest.fixture
@@ -15,9 +15,7 @@ def vctk_dataset():
         test_clean="clean_testset_wav",
         test_noisy="noisy_testset_wav",
     )
-    dataset = MayaDataset(
+    dataset = EnhancerDataset(name="vctk", root_dir=root_dir, files=files)
-        name="vctk", root_dir=root_dir, files=files, sampling_rate=16000
-    )
     return dataset