This is an unofficial implentation of PHASEN: A Phase-and-Harmonics-Aware Speech Enhancement Network.
You could download datasets from https://datashare.is.ed.ac.uk/handle/10283/1942.
./tools/resample.py could be used to resample .wav file, and ./tools/clip.py could be used to clip wave segment into 1 second. Before training model, make sure the datasets has been resampled to 16kHz and cliped into 1 second.
The model implentation is in ./phasen.py.
You could use ./run.py to train and test. The model parameters will be stored in directory ./model.
Before you use any .py file in this repo, make sure you have changed the parameters in that file, like epochs, learning_rate, etc.
- Python
- Pytorch
- Librosa
- Numpy
- Soundfile