Home

Awesome

VoiceFilter

Note from Seung-won (2020.10.25)

Hi everyone! It's Seung-won from MINDs Lab, Inc. It's been a long time since I've released this open-source, and I didn't expect this repository to grab such a great amount of attention for a long time. I would like to thank everyone for giving such attention, and also Mr. Quan Wang (the first author of the VoiceFilter paper) for referring this project in his paper.

Actually, this project was done by me when it was only 3 months after I started studying deep learning & speech separation without a supervisor in the relevant field. Back then, I didn't know what is a power-law compression, and the correct way to validate/test the models. Now that I've spent more time on deep learning & speech since then (I also wrote a paper published at Interspeech 2020 😊), I can observe some obvious mistakes that I've made. Those issues were kindly raised by GitHub users; please refer to the Issues and Pull Requests for that. That being said, this repository can be quite unreliable, and I would like to remind everyone to use this code at their own risk (as specified in LICENSE).

Unfortunately, I can't afford extra time on revising this project or reviewing the Issues / Pull Requests. Instead, I would like to offer some pointers to newer, more reliable resources:

Thanks for reading, and I wish everyone good health during the global pandemic situation.

Best regards, Seung-won Park

Unofficial PyTorch implementation of Google AI's: VoiceFilter: Targeted Voice Separation by Speaker-Conditioned Spectrogram Masking.

Result

Audio Sample

Metric

Median SDRPaperOurs
before VoiceFilter2.51.9
after VoiceFilter12.610.2

Dependencies

  1. Python and packages

    This code was tested on Python 3.6 with PyTorch 1.0.1. Other packages can be installed by:

    pip install -r requirements.txt

  2. Miscellaneous

    ffmpeg-normalize is used for resampling and normalizing wav files. See README.md of ffmpeg-normalize for installation.

Prepare Dataset

  1. Download LibriSpeech dataset

    To replicate VoiceFilter paper, get LibriSpeech dataset at http://www.openslr.org/12/. train-clear-100.tar.gz(6.3G) contains speech of 252 speakers, and train-clear-360.tar.gz(23G) contains 922 speakers. You may use either, but the more speakers you have in dataset, the more better VoiceFilter will be.

  2. Resample & Normalize wav files

    First, unzip tar.gz file to desired folder:

    tar -xvzf train-clear-360.tar.gz

    Next, copy utils/normalize-resample.sh to root directory of unzipped data folder. Then:

    vim normalize-resample.sh # set "N" as your CPU core number.
chmod a+x normalize-resample.sh
./normalize-resample.sh # this may take long

  3. Edit config.yaml

    cd config
cp default.yaml config.yaml
vim config.yaml

  4. Preprocess wav files

    In order to boost training speed, perform STFT for each files before training by:

    python generator.py -c [config yaml] -d [data directory] -o [output directory] -p [processes to run]

    This will create 100,000(train) + 1000(test) data. (About 160G)

Train VoiceFilter

  1. Get pretrained model for speaker recognition system

    VoiceFilter utilizes speaker recognition system (d-vector embeddings). Here, we provide pretrained model for obtaining d-vector embeddings.

    This model was trained with VoxCeleb2 dataset, where utterances are randomly fit to time length [70, 90] frames. Tests are done with window 80 / hop 40 and have shown equal error rate about 1%. Data used for test were selected from first 8 speakers of VoxCeleb1 test dataset, where 10 utterances per each speakers are randomly selected.

    Update: Evaluation on VoxCeleb1 selected pair showed 7.4% EER.

    The model can be downloaded at this GDrive link.

  2. Run

    After specifying train_dir, test_dir at config.yaml, run:

    python trainer.py -c [config yaml] -e [path of embedder pt file] -m [name]

    This will create chkpt/name and logs/name at base directory(-b option, . in default)

  3. View tensorboardX

    tensorboard --logdir ./logs

  4. Resuming from checkpoint

    python trainer.py -c [config yaml] --checkpoint_path [chkpt/name/chkpt_{step}.pt] -e [path of embedder pt file] -m name

Evaluate

python inference.py -c [config yaml] -e [path of embedder pt file] --checkpoint_path [path of chkpt pt file] -m [path of mixed wav file] -r [path of reference wav file] -o [output directory]

Possible improvments

Author

Seungwon Park at MINDsLab (yyyyy@snu.ac.kr, swpark@mindslab.ai)

License

Apache License 2.0

This repository contains codes adapted/copied from the followings: