Share AI Models and Methods


DAM-Softmax loss is an improvement of AM-Softmax, it is proposed in paper: Dynamic Margin Softmax Loss for Speaker Verification.

DAM-Softmax Visualization

This loss function is formulated as:

DAM-Softmax Loss

From this paper, we can find:

m = 0.2 ,λ = 2, s = 30

DAM-Softmax vs AM-Softmax

We can find the different performance between them in speaker verification task.

DAM-Softmax vs AM-Softmax

We should notice : in this paper, m = 0.2 and s = 30 in AM-Softmax