Hugging Face's logo

mechanicalsea
/
efficient-tdnn

PyTorch
English
embeddings
Speaker
Verification
Identification
NAS
TDNN
Model card Files Community
efficient-tdnn / README.md
mechanicalsea's picture
mechanicalsea
update readme and git attr
bf864fd
preview code
|
raw history blame
No virus
2.23 kB
metadata
language:
  - en
license: mit
tags:
  - embeddings
  - Speaker
  - Verification
  - Identification
  - NAS
  - TDNN
  - pytorch
datasets:
  - voxceleb1
  - voxceleb2
metrics:
  - EER
  - minDCF:
      - p_target: 0.01

EfficientTDNN

Model Version are listed as follows.

  • Dynamic Kernel: The model enables various kernel sizes in {1,3,5}, kernel/kernel.torchparams.
  • Dynamic Depth: The model enables additional various depth in {2,3,4} based on Dynamic Kernel version, depth/depth.torchparams.
  • Dynamic Width 1: The model enable additional various width in [0.5, 1.0] based on Dynamic Depth version, width1/width1.torchparams.
  • Dynamic Width 2: The model enable additional various width in [0.25, 0.5] based on Dynamic Width 1 version, width2/width2.torchparams.

Furthermore, some subnets are given in the form of the weights of batchnorm corresponding to their trained supernets as follows.

  • Dynamic Kernel
    1. kernel/kernel.max.bn.tar
    2. kernel/kernel.Kmin.bn.tar
  • Dynamic Depth
    1. depth/depth.max.bn.tar
    2. depth/depth.Kmin.bn.tar
    3. depth/depth.Dmin.bn.tar
    4. depth/depth.3.512.5.5.3.3.1536.bn.tar
    5. depth/depth.ecapa-tdnn.3.512.512.512.512.5.3.3.3.1536.bn.tar
  • Dynamic Width 1
    1. width1/width1.torchparams
    2. width1/width1.max.bn.tar
    3. width1/width1.Kmin.bn.tar
    4. width1/width1.Dmin.bn.tar
    5. width1/width1.C1min.bn.tar
    6. width1/width1.3.383.256.256.256.5.3.3.3.768.bn.tar
  • Dynamic Width 2
    1. width2/width2.max.bn.tar
    2. width2/width2.Kmin.bn.tar
    3. width2/width2.Dmin.bn.tar
    4. width2/width2.C1min.bn.tar
    5. width2/width2.C2min.bn.tar
    6. width2/width2.3.384.3.1152.bn.tar
    7. width2/width2.3.256.256.384.384.1.3.5.3.1152.bn.tar
    8. width2/width2.2.256.256.256.3.3.3.400.bn.tar

The tag is described as follows.

  • max: (4, [512, 512, 512, 512, 512], [5, 5, 5, 5, 5], 1536)
  • Kmin: (4, [512, 512, 512, 512, 512], [1, 1, 1, 1, 1], 1536)
  • Dmin: (2, [512, 512, 512], [1, 1, 1], 1536)
  • C1min: (2, [256, 256, 256], [1, 1, 1], 768)
  • C2min: (2, [128, 128, 128], [1, 1, 1], 384)

More details about EfficentTDNN can be found in the paper EfficientTDNN.