Make simple inference from #1 with sample_rate configurable

.gitignore

@@ -0,0 +1,160 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/

README.md

@@ -11,14 +11,17 @@ from various objective and subjective evaluation metrics. Sound demos can be fou
 
 ## Datasets
 
-- [Microsoft DNS 2020](https://arxiv.org/ftp/arxiv/papers/2005/2005.13981.pdf) dataset. The dataset, pre-processing codes, and instruction to generate training data can be found in [this link](https://github.com/microsoft/DNS-Challenge/tree/interspeech2020/master). Assume the dataset is stored under ```./dns```. Before generating clean-noisy data pairs, modify the following parameters in their ```noisyspeech_synthesizer.cfg``` file: 
+- [Microsoft DNS 2020](https://arxiv.org/ftp/arxiv/papers/2005/2005.13981.pdf) dataset. The dataset, pre-processing codes, and instruction to generate training data can be found in [this link](https://github.com/microsoft/DNS-Challenge/tree/interspeech2020/master). Assume the dataset is stored under ``./dns``. Before generating clean-noisy data pairs, modify the following parameters in their ``noisyspeech_synthesizer.cfg`` file:
+
 ```
-total_hours: 500, 
-snr_lower: -5, 
-snr_upper: 25, 
+total_hours: 500,
+snr_lower: -5,
+snr_upper: 25,
 total_snrlevels: 31
 ```
+
 And also update paths as (since their original code uses Windows-style paths)
+
 ```
 noise_dir: ./datasets/noise
 speech_dir: ./datasets/clean
@@ -28,25 +31,32 @@ noise_destination: ./training_set/noise
 log_dir: ./logs
 unit_tests_log_dir: ./unittests_logs
 ```
-Then, for conciseness and to comply with our data loading codes, modify file names (lines 198-201) in their ```noisyspeech_synthesizer_singleprocess.py``` to 
+
+Then, for conciseness and to comply with our data loading codes, modify file names (lines 198-201) in their ``noisyspeech_synthesizer_singleprocess.py`` to
+
 ```
 noisyfilename = 'fileid_' + str(file_num) + '.wav'
 cleanfilename = 'fileid_' + str(file_num) + '.wav'
 noisefilename = 'fileid_' + str(file_num) + '.wav'
 ```
-To generate training data, run 
+
+To generate training data, run
+
 ```
 python noisyspeech_synthesizer_singleprocess.py
 ```
+
 It is also recommended to rename files in the test set for conciseness:
+
 ```
 cd ./dns/datasets/test_set/synthetic/no_reverb/noisy/
-for NAME in $(ls ./); do arr=(${NAME//fileid_/ }); mv ${NAME} noisy_fileid_${arr[1]}; done 
+for NAME in $(ls ./); do arr=(${NAME//fileid_/ }); mv ${NAME} noisy_fileid_${arr[1]}; done
 ```
 
 After these steps, we assume that the structure of the dataset folder is:
+
 ```
-Training sets: 
+Training sets:
 ./dns/training_set/clean/fileid_{0..59999}.wav
 ./dns/training_set/noisy/fileid_{0..59999}.wav
 ./dns/training_set/noise/fileid_{0..59999}.wav
@@ -56,49 +66,57 @@ Testing sets (no-reverb):
 ./dns/datasets/test_set/synthetic/no_reverb/noisy/noisy_fileid_{0..299}.wav
 ```
 
-- Other datasets are also supported; lines 49-50 of ```dataset.py``` need to be carefully changed to handle paths and file names.
+- Other datasets are also supported; lines 49-50 of ``dataset.py`` need to be carefully changed to handle paths and file names.
 
 ## Training
 
-The ```$EXP``` variable can be any config name in ```./configs/```, such as ```DNS-large-full``` and ```DNS-large-high```. The default experiment path is ```./exp```; it can be changed by modifying ```train_config[log[directory]]``` in the config files. ```trainset_config[root]``` needs to be set as the root path of the dataset. Then, the training code is
+The ``$EXP`` variable can be any config name in ``./configs/``, such as ``DNS-large-full`` and ``DNS-large-high``. The default experiment path is ``./exp``; it can be changed by modifying ``train_config[log[directory]]`` in the config files. ``trainset_config[root]`` needs to be set as the root path of the dataset. Then, the training code is
 
-```python3 distributed.py -c configs/${EXP}.json```
+``python3 distributed.py -c configs/${EXP}.json``
 
 We use 8 GPUs for training. The global batch size is 64 and we train the models for 250K iterations. Note that, this is different from the training setup in our paper i.e., 1M iterations with a batch size of 16. We find negligible difference in terms of objective and subjective evaluation, but the current setup is faster.
 
-**Pre-trained** models for denoising are provided in ```./exp/${EXP}/checkpoint/pretrained.pkl``` (each one has size ~177Mb; use ```git lfs``` to download). Note that these models are not trained to remove reverb. 
+**Pre-trained** models for denoising are provided in ``./exp/${EXP}/checkpoint/pretrained.pkl`` (each one has size ~177Mb; use ``git lfs`` to download). Note that these models are not trained to remove reverb.
 
 ## Denoising
 
-We perform denoising on the DNS no-reverb test dataset. The output path is ```gen_config[output_directory]```, which is ```./exp``` by default. The denoising code is
+We perform denoising on the DNS no-reverb test dataset. The output path is ``gen_config[output_directory]``, which is ``./exp`` by default. The denoising code is
 
-```python denoise.py -c configs/${EXP}.json --ckpt_iter ${ITERATION}```
+``python denoise.py -c configs/${EXP}.json --ckpt_iter ${ITERATION}``
 
 For example, if you want to use pre-trained models to denoise, run:
 
-```python denoise.py -c configs/DNS-large-high.json --ckpt_iter pretrained```
+``python denoise.py -c configs/DNS-large-high.json --ckpt_iter pretrained``
 
 1 GPU is used for denoising.
 
+## Simple Denoising without dataset
+
+There is a script `simple_infrence.py` to denoise single or multiple files. It also does automatic fp16 conversion. You can also adjust batch size to fit your vram.
+
+```denoise_simple.py python denoise.py -c configs/${EXP}.json --ckpt_path ${file.plk} -b 1000000 -sr 16000 \<you file list here>````
+
+``` python denoise_simple.py -c configs/DNS-large-high.json --ckpt_pat ./exp/DNS-large-high/checkpoint/pretrained.pkl ./test.mp4```
+
 ## Evaluation
 
 The following evaluation code generates [PESQ](https://www.itu.int/rec/T-REC-P.862) and [STOI](https://ceestaal.nl/code/) scores. More evaluation metrics can be found in the [SEGAN (PyTorch)](https://github.com/santi-pdp/segan_pytorch) repo.
 
-```python python_eval.py -d dns -e ${PATH_TO_DENOISED_SPEECH} -t ${PATH_TO_TESTSET_PATH} >> eval.log```
+``python python_eval.py -d dns -e ${PATH_TO_DENOISED_SPEECH} -t ${PATH_TO_TESTSET_PATH} >> eval.log``
 
 1 GPU is used for evaluation.
 
 ## Requirements
 
 To synthesize [Microsoft DNS 2020](https://arxiv.org/ftp/arxiv/papers/2005/2005.13981.pdf) training data, you need [these dependencies](https://github.com/microsoft/DNS-Challenge/blob/interspeech2020/master/requirements.txt). If you just want to evaluate our pre-trained models on the test data, you may jump this.
 
-Our code is tested on 8 NVIDIA V100 GPUs. You need to install very standard dependencies: ```numpy``` and ```scipy``` for scientific computing, ```torch, torchvision, torchaudio``` for deep learning and data loading, ```pesq, pystoi``` for audio evaluation, and ```tqdm``` for visualization.
+Our code is tested on 8 NVIDIA V100 GPUs. You need to install very standard dependencies: ``numpy`` and ``scipy`` for scientific computing, ``torch, torchvision, torchaudio`` for deep learning and data loading, ``pesq, pystoi`` for audio evaluation, and ``tqdm`` for visualization.
 
 ## References
 
-The code structure and distributed training are adapted from [WaveGlow (PyTorch)](https://github.com/NVIDIA/waveglow) (BSD-3-Clause license). The ```stft_loss.py``` is adapted from [ParallelWaveGAN (PyTorch)](https://github.com/kan-bayashi/ParallelWaveGAN) (MIT license). The self-attention blocks in ```network.py``` is adapted from [Attention is all you need (PyTorch)](https://github.com/jadore801120/attention-is-all-you-need-pytorch) (MIT license), which borrows from [OpenNMT-py](https://github.com/OpenNMT/OpenNMT-py) (MIT license). The learning rate scheduler in ```util.py``` is adapted from [VQVAE2 (PyTorch)](https://github.com/rosinality/vq-vae-2-pytorch) (MIT license). Some utility functions are borrowed from [DiffWave (PyTorch)](https://github.com/philsyn/DiffWave-Vocoder) (MIT license) and [WaveGlow (PyTorch)](https://github.com/NVIDIA/waveglow) (BSD-3-Clause license).
+The code structure and distributed training are adapted from [WaveGlow (PyTorch)](https://github.com/NVIDIA/waveglow) (BSD-3-Clause license). The ``stft_loss.py`` is adapted from [ParallelWaveGAN (PyTorch)](https://github.com/kan-bayashi/ParallelWaveGAN) (MIT license). The self-attention blocks in ``network.py`` is adapted from [Attention is all you need (PyTorch)](https://github.com/jadore801120/attention-is-all-you-need-pytorch) (MIT license), which borrows from [OpenNMT-py](https://github.com/OpenNMT/OpenNMT-py) (MIT license). The learning rate scheduler in ``util.py`` is adapted from [VQVAE2 (PyTorch)](https://github.com/rosinality/vq-vae-2-pytorch) (MIT license). Some utility functions are borrowed from [DiffWave (PyTorch)](https://github.com/philsyn/DiffWave-Vocoder) (MIT license) and [WaveGlow (PyTorch)](https://github.com/NVIDIA/waveglow) (BSD-3-Clause license).
 
-For more evaluation methods, we refer readers to look at [SEGAN (PyTorch)](https://github.com/santi-pdp/segan_pytorch/blob/master/segan/utils.py) (MIT license). For more data augmentation methods, we refer readers to look at [FAIR-denoiser](https://github.com/facebookresearch/denoiser/blob/main/denoiser/augment.py) (CC-BY-NC 4.0 license). 
+For more evaluation methods, we refer readers to look at [SEGAN (PyTorch)](https://github.com/santi-pdp/segan_pytorch/blob/master/segan/utils.py) (MIT license). For more data augmentation methods, we refer readers to look at [FAIR-denoiser](https://github.com/facebookresearch/denoiser/blob/main/denoiser/augment.py) (CC-BY-NC 4.0 license).
 
 ## Citation