Skip to main content

Table 1 The structure of RawNet-SA

From: Anchor voiceprint recognition in live streaming via RawNet-SA and gated recurrent unit

Layer Input:59049 Output
Sinc Sinc (251,1,128)
MaxPool (3)
BN
LeakyReLU
(19,683,128)
Resblock×3 BN (725,128)
LeakyReLU
Conv (3,1,128)
BN
LeakyReLU
Conv (3,1,128)
MaxPool (3)
FMS
Resblock × 3 BN (26,256)
LeakyReLU
Conv (3,1,256)
BN
LeakyReLU
Conv (3,1,256)
MaxPool (3)
SA
Aggregation GRU (1024) (1024)
Embedding FC (1024) (1024)