PyTorch implementation ASR frontend, like PCEN, Mel filter bank log energy.
The following is a example for using PCEN:
import pcen
import numpy as np
b, s, d = 32, 100, 40
filterbanks = np.random.uniform(low=0.5, high=13.3, size=(b, s, d))
filterbanks = torch.from_numpy(filterbanks.astype(dtype=np.float32))
trainable_pcen = pcen.Pcen(d)
pcen_features = trainable_pcen(filterbanks)
Wang, Yuxuan, Pascal Getreuer, Thad Hughes, Richard F. Lyon, and Rif A. Saurous. Trainable frontend for robust and far-field keyword spotting. In Acoustics, Speech and Signal Processing (ICASSP), 2017 IEEE International Conference on, pp. 5670-5674. IEEE, 2017.
@inproceedings{wang2017trainable,
title={Trainable frontend for robust and far-field keyword spotting},
author={Wang, Yuxuan and Getreuer, Pascal and Hughes, Thad and Lyon, Richard F and Saurous, Rif A},
booktitle={Acoustics, Speech and Signal Processing (ICASSP), 2017 IEEE International Conference on},
pages={5670--5674},
year={2017},
organization={IEEE}
}