WenwanChen / pvad

speaker conditioned voice activity detection

pvad

speaker conditioned voice activity detection replicated from https://arxiv.org/abs/1908.04284

Classifier: {non-speech, target speaker, and non-target speaker}

1. Synthetic dataset generation
   prep4kaldi.sh
   flac_to_wav.sh
   concat.sh concat.py
   augment.py

2. Prepare target speaker embeddings
   extract_embeddings.py

3. Extract features and labels
   correct_target_labels.py
   fbank.py
   feature_labels.py

4. Data loader
   dataloader.py
   dataloader_test.py

5. Model definition and traning
   pvad_training.py

6. Saved model
   checkpoint_oct22_coswarm.t7

7. Test
   test.py