This repo is a TensorFlow implementation of Griffin-Lim algorithm for voice reconstruction.
For comparision, a librosa implementation version is also included in this repo.
- TensorFlow >=1.3
- Python 3.x
- scipy
- numpy
- librosa
To test the performance, we could execute the following command:
python test.py
The test.py
script does the following things in order:
- Extract the spectrogram features from waves
- Reconstruction voice using Griffin-lim algorithm
All data used is in the data
folder. If we want to use other data, replace it.
Some code are borrowed from the following repo: