This repository implements the following paper:
TitaNet: Neural Model for speaker representation with 1D Depth-wise separable convolutions and global context,
Nithin Rao Koluguri, Taejin Park, Boris Ginsburg,
https://ieeexplore.ieee.org/document/9746806.
Data training uses from open source Vietnamese Speaker Verification. It contains about 25 hours audios, which are crawled from 1015 vietnamese youtube channels. This data is not verified by human, so it's quality is low
Data testing is took from VLSP 2021, task Speaker verification. The data is built from VLSP 2020, and crawled from youtube. Each sample includes a pair audios, and label is it the same speaker. Dataset is split into 3 parts
- Public dataset: Most samples are taken from VLSP 2020 dataset, which is clean
- Private dataset 1: Samples are taken from youtube audios, which have varied background environments including inaudible chatter, laughs, street noise, school, music,....
- Private dataset 2: It includes most hard samples, for testing robustness of systems.
Use Equal Error Rate (EER) metric
| Dataset | EER |
|---|---|
| Public dataset | 7.36% |
| Private dataset 1 | 7.92% |
| Private dataset 2 | 9.76% |
TitaNet: Neural Model for speaker representation with 1D Depth-wise separable convolutions and global context,
Nithin Rao Koluguri, Taejin Park, Boris Ginsburg,
https://ieeexplore.ieee.org/document/9746806.
VLSP 2021 - SV challenge: Vietnamese Speaker Verification
in Noisy Environments
https://jcsce.vnu.edu.vn/index.php/jcsce/article/view/333
Vietnamese Speaker Verification
https://www.kaggle.com/datasets/dungnasa10/vietnamese-speaker-verification