Densely Connected Time Delay Neural Network

PyTorch implementation of Densely Connected Time Delay Neural Network (D-TDNN) in our paper "Densely Connected Time Delay Neural Network for Speaker Verification" (INTERSPEECH 2020).

News

• [2023-05-04] 3D-Speaker supports training of CAM++ model and can be easily extended to support training of raw D-TDNN and CAM models. They also released a Chinese speaker embedding model trained on 200k speakers and an English speaker embedding model trained on VoxCeleb.

• [2023-03-04] CAM++ achieved superior performance with lower computational complexity and faster inference speed than popular ECAPA-TDNN and ResNet34 systems.

  H. Wang, S. Zheng, Y. Chen, L. Cheng, and Q. Chen, "CAM++: A Fast and Efficient Network for Speaker Verification Using Context-Aware Masking"

  	VoxCeleb1-E	VoxCeleb1-H	CN-Celeb
  ECAPA-TDNN	1.07/0.1185	1.98/0.1956	7.45/0.4127
  D-TDNN	1.63/0.1748	2.86/0.2571	8.41/0.4683
  CAM	1.18/0.1257*	2.15/0.1966*	-
  CAM++	0.89/0.0995	1.76/0.1729	6.78/0.3830

• [2021-09-05] TimeDelay is replaced by Conv1d by default, since convolution is better optimized in all kinds of deep learning frameworks (Note: The pretrained models are directly converted from the old ones so that the results might be slightly different from those in the paper).

• [2021-08-28] D-TDNN and D-TDNN-SS outperform SOTA system on the AP20-OLR-dialect-task of oriental language recognition (OLR) challenge 2020 (WeChat artical / paper), showing their potential on other speech processing tasks.

• [2021-02-01] CAM adopts D-TDNN backbone and is enhanced by context-aware masking.

  Y.-Q. Yu, S. Zheng, H. Suo, Y. Lei, and W.-J. Li, "CAM: Context-Aware Masking for Robust Speaker Verification" (ICASSP 2021)

  	VoxCeleb1-E	VoxCeleb1-H
  CAM	1.18/0.1257	2.15/0.1966

Pretrained Models

We provide the pretrained models which can be used in many tasks such as:

• Speaker Verification
• Speaker-Dependent Speech Separation
• Multi-Speaker Text-to-Speech
• Voice Conversion

Usage

Data preparation

You can either use Kaldi toolkit:

• Download VoxCeleb1 test set and unzip it.
• Place prepare_voxceleb1_test.sh under $kaldi_root/egs/voxceleb/v2 and change the $datadir and $voxceleb1_root in it.
• Run chmod +x prepare_voxceleb1_test.sh && ./prepare_voxceleb1_test.sh to generate 30-dim MFCCs.
• Place the trials under $datadir/test_no_sil.

Or checkout the kaldifeat branch if you do not want to install Kaldi.

Test

• Download the pretrained D-TDNN model and run:

python evaluate.py --root $datadir/test_no_sil --model D-TDNN --checkpoint dtdnn.pth --device cuda

Evaluation

VoxCeleb1-O

Citation

If you find D-TDNN helps your research, please cite

@inproceedings{DBLP:conf/interspeech/YuL20,
  author    = {Ya-Qi Yu and
               Wu-Jun Li},
  title     = {Densely Connected Time Delay Neural Network for Speaker Verification},
  booktitle = {Annual Conference of the International Speech Communication Association (INTERSPEECH)},
  pages     = {921--925},
  year      = {2020}
}

Revision of the Paper

References:

[16] X. Li, W. Wang, X. Hu, and J. Yang, "Selective Kernel Networks," in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 510-519.