Publication Details
Investigation of Specaugment for Deep Speaker Embedding Learning
Rohdin Johan A., Dr. (DCGM FIT BUT)
Plchot Oldřich, Ing., Ph.D. (DCGM FIT BUT)
Burget Lukáš, doc. Ing., Ph.D. (DCGM FIT BUT)
Yu Kai (SJTU)
Černocký Jan, prof. Dr. Ing. (DCGM FIT BUT)
speaker embedding, on-the-fly data augmentation, speaker verification, specaugment
SpecAugment is a newly proposed data augmentation method for speech recognition. By randomly masking bands in the log Mel spectogram this method leads to impressive performance improvements. In this paper, we investigate the usage of SpecAugment for speaker verification tasks. Two different models, namely 1-D convolutional TDNN and 2-D convolutional ResNet34, trained with either Softmax or AAM-Softmax loss, are used to analyze SpecAugments effectiveness. Experiments are carried out on the Voxceleb and NIST SRE 2016 dataset. By applying SpecAugment to the original clean data in an on-the-fly manner without complex off-line data augmentation methods, we obtained 3.72% and 11.49% EER for NIST SRE 2016 Cantonese and Tagalog, respectively. For Voxceleb1 evaluation set, we obtained 1.47% EER.
@INPROCEEDINGS{FITPUB12278,
author = "Shuai Wang and A. Johan Rohdin and Old\v{r}ich Plchot and Luk\'{a}\v{s} Burget and Kai Yu and Jan \v{C}ernock\'{y}",
title = "Investigation of Specaugment for Deep Speaker Embedding Learning",
pages = "7139--7143",
booktitle = "ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings",
year = 2020,
location = "Barcelona, ES",
publisher = "IEEE Signal Processing Society",
ISBN = "978-1-5090-6631-5",
doi = "10.1109/ICASSP40776.2020.9053481",
language = "english",
url = "https://www.fit.vut.cz/research/publication/12278"
}