Source code of the paper Multi-Layered Cepstrum for Instantaneous Frequency Estimation accepted at GlobalSIP 2018.
- Numpy
- Scipy
- Torch (optional for faster performance)
- Librosa (file IO)
- Matplotlib (visualization)
- Pypianoroll (visualization)
- Donwload bach10 dataset.
- Run with default parameters.
python bach10.py your/download/path/01-AchGottundHerr/01-AchGottundHerr.wav \
--f0_file your/download/path/01-AchGottundHerr/01-AchGottundHerr-GTF0s.mat
.
.
.
Time cost: 5.6769 seconds
Precision: 0.7661, Recall: 0.9170, F-score:, 0.8348
We also implement a faster version using PyTorch as torch_bach10.py
, and it can run roughly 2 times faster on CPU.
Add --cuda
can further utilize computational resources on GPU.