Code to generate modgdgram as a feature for musical source separation, speech separation and speech enhancement The code receives input wavefile with specifications of FFT points, window length, window shift and the sampling rate. The output produces a modgdgram, which can be fed to a DRNN module instead of the magnitude spectrogram.
The details of the work are described in the website https://sites.google.com/site/groupdelayfeatureformusicbss/ and paper https://ieeexplore.ieee.org/document/7746672
If you use this code for feature extraction, please cite the following paper:
J. Sebastian and H. A. Murthy, "Group delay based music source separation using deep recurrent neural networks," 2016 International Conference on Signal Processing and Communications (SPCOM), Bangalore, 2016, pp. 1-5, doi: 10.1109/SPCOM.2016.7746672.