Data Augmentation with ECAPA-TDNN Architecture for Automatic Speaker Recognition

Pinyan Li; Lap Man Hoi; Yapeng Wang; Sio Kei Im

doi:10.1109/ICRERA59003.2023.10269366

Data Augmentation with ECAPA-TDNN Architecture for Automatic Speaker Recognition

Faculty of Applied Sciences

Macao Polytechnic University

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

2 Citations (Scopus)

Abstract

This paper focuses on seven data augmentation methods based on the Emphasized Channel Attention Propagation and Aggregation-Time Delay Neural Network (ECAPA-TDNN) model for increasing the diversity of training data to improve model accuracy and true positive rate (TPR/recall). We propose a method to improve classification performance by replacing and reducing the datasets. We also verified the effect of the number of layers on the classification performance by modifying the number of layers of the SE-Res2Block in the ECAPA-TDNN model. The proposed method is validated with the ZhVoice and VoxCeleb datasets, and the results show that the best model accuracy and classification performance can be obtained by using ZhVoice with seven data augmentations on a 3-layer SE-Res2Block. The accuracy reached 0.9477, the TPR reached 0.8945, and the EER was 0.1278. We also used the diagonal cosine algorithm to determine the similarity between two speakers, validating the classification performance of the model.

Original language	English
Title of host publication	12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	414-420
Number of pages	7
ISBN (Electronic)	9798350337938
DOIs	https://doi.org/10.1109/ICRERA59003.2023.10269366
Publication status	Published - 2023
Event	12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023 - Oshawa, Canada Duration: 29 Aug 2023 → 1 Sept 2023

Publication series

Name	12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023

Conference

Conference	12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023
Country/Territory	Canada
City	Oshawa
Period	29/08/23 → 1/09/23

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 7 Affordable and Clean Energy

Keywords

ECAPA-TDNN
automatic speaker recognition
data augmentation

Access to Document

10.1109/ICRERA59003.2023.10269366

Cite this

Li, P., Hoi, L. M., Wang, Y., & Im, S. K. (2023). Data Augmentation with ECAPA-TDNN Architecture for Automatic Speaker Recognition. In 12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023 (pp. 414-420). (12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICRERA59003.2023.10269366

Li, Pinyan ; Hoi, Lap Man ; Wang, Yapeng et al. / Data Augmentation with ECAPA-TDNN Architecture for Automatic Speaker Recognition. 12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023. Institute of Electrical and Electronics Engineers Inc., 2023. pp. 414-420 (12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023).

@inproceedings{384cd897411a40569937c5b8b1b97aae,

title = "Data Augmentation with ECAPA-TDNN Architecture for Automatic Speaker Recognition",

abstract = "This paper focuses on seven data augmentation methods based on the Emphasized Channel Attention Propagation and Aggregation-Time Delay Neural Network (ECAPA-TDNN) model for increasing the diversity of training data to improve model accuracy and true positive rate (TPR/recall). We propose a method to improve classification performance by replacing and reducing the datasets. We also verified the effect of the number of layers on the classification performance by modifying the number of layers of the SE-Res2Block in the ECAPA-TDNN model. The proposed method is validated with the ZhVoice and VoxCeleb datasets, and the results show that the best model accuracy and classification performance can be obtained by using ZhVoice with seven data augmentations on a 3-layer SE-Res2Block. The accuracy reached 0.9477, the TPR reached 0.8945, and the EER was 0.1278. We also used the diagonal cosine algorithm to determine the similarity between two speakers, validating the classification performance of the model.",

keywords = "ECAPA-TDNN, automatic speaker recognition, data augmentation",

author = "Pinyan Li and Hoi, \{Lap Man\} and Yapeng Wang and Im, \{Sio Kei\}",

note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; 12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023 ; Conference date: 29-08-2023 Through 01-09-2023",

year = "2023",

doi = "10.1109/ICRERA59003.2023.10269366",

language = "English",

series = "12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "414--420",

booktitle = "12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023",

address = "United States",

}

Li, P, Hoi, LM , Wang, Y & Im, SK 2023, Data Augmentation with ECAPA-TDNN Architecture for Automatic Speaker Recognition. in 12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023. 12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023, Institute of Electrical and Electronics Engineers Inc., pp. 414-420, 12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023, Oshawa, Canada, 29/08/23. https://doi.org/10.1109/ICRERA59003.2023.10269366

Data Augmentation with ECAPA-TDNN Architecture for Automatic Speaker Recognition. / Li, Pinyan; Hoi, Lap Man ; Wang, Yapeng et al.
12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023. Institute of Electrical and Electronics Engineers Inc., 2023. p. 414-420 (12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Data Augmentation with ECAPA-TDNN Architecture for Automatic Speaker Recognition

AU - Li, Pinyan

AU - Hoi, Lap Man

AU - Wang, Yapeng

AU - Im, Sio Kei

PY - 2023

Y1 - 2023

N2 - This paper focuses on seven data augmentation methods based on the Emphasized Channel Attention Propagation and Aggregation-Time Delay Neural Network (ECAPA-TDNN) model for increasing the diversity of training data to improve model accuracy and true positive rate (TPR/recall). We propose a method to improve classification performance by replacing and reducing the datasets. We also verified the effect of the number of layers on the classification performance by modifying the number of layers of the SE-Res2Block in the ECAPA-TDNN model. The proposed method is validated with the ZhVoice and VoxCeleb datasets, and the results show that the best model accuracy and classification performance can be obtained by using ZhVoice with seven data augmentations on a 3-layer SE-Res2Block. The accuracy reached 0.9477, the TPR reached 0.8945, and the EER was 0.1278. We also used the diagonal cosine algorithm to determine the similarity between two speakers, validating the classification performance of the model.

AB - This paper focuses on seven data augmentation methods based on the Emphasized Channel Attention Propagation and Aggregation-Time Delay Neural Network (ECAPA-TDNN) model for increasing the diversity of training data to improve model accuracy and true positive rate (TPR/recall). We propose a method to improve classification performance by replacing and reducing the datasets. We also verified the effect of the number of layers on the classification performance by modifying the number of layers of the SE-Res2Block in the ECAPA-TDNN model. The proposed method is validated with the ZhVoice and VoxCeleb datasets, and the results show that the best model accuracy and classification performance can be obtained by using ZhVoice with seven data augmentations on a 3-layer SE-Res2Block. The accuracy reached 0.9477, the TPR reached 0.8945, and the EER was 0.1278. We also used the diagonal cosine algorithm to determine the similarity between two speakers, validating the classification performance of the model.

KW - ECAPA-TDNN

KW - automatic speaker recognition

KW - data augmentation

UR - https://www.scopus.com/pages/publications/85175647550

U2 - 10.1109/ICRERA59003.2023.10269366

DO - 10.1109/ICRERA59003.2023.10269366

M3 - Conference contribution

AN - SCOPUS:85175647550

T3 - 12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023

SP - 414

EP - 420

BT - 12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023

Y2 - 29 August 2023 through 1 September 2023

ER -

Li P, Hoi LM , Wang Y , Im SK. Data Augmentation with ECAPA-TDNN Architecture for Automatic Speaker Recognition. In 12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023. Institute of Electrical and Electronics Engineers Inc. 2023. p. 414-420. (12th IEEE International Conference on Renewable Energy Research and Applications, ICRERA 2023). doi: 10.1109/ICRERA59003.2023.10269366

Data Augmentation with ECAPA-TDNN Architecture for Automatic Speaker Recognition

Abstract

Publication series

Conference

UN SDGs

Keywords

Access to Document

Other files and links

Fingerprint

Cite this