new definition of embeddings

nnet/xvector.py

@@ -98,14 +98,14 @@ class Xtractor(torch.nn.Module):
         seg_emb_0 = torch.cat([mean, std], dim=1)
         # No batch-normalisation after this layer
         seg_emb_1 = self.dropout_lin0(seg_emb_0)
-        seg_emb_1 = self.activation(self.seg_lin0(seg_emb_1))
+        seg_emb_2 = self.activation(self.seg_lin0(seg_emb_1))
         # new layer with batch Normalization
-        seg_emb_2 = self.dropout_lin1(seg_emb_1)
-        seg_emb_3 = self.norm6(self.activation(self.seg_lin1(seg_emb_2)))
+        seg_emb_3 = self.dropout_lin1(seg_emb_2)
+        seg_emb_4 = self.norm6(self.activation(self.seg_lin1(seg_emb_3)))
         # No batch-normalisation after this layer
-        seg_emb_4 = self.activation(self.seg_lin2(seg_emb_3))
-        #seg_emb_3 = self.seg_lin2(seg_emb_2)
-        return seg_emb_4
+        seg_emb_5 = self.activation(self.seg_lin2(seg_emb_4))
+        seg_output = torch.nn.functional.softmax(seg_emb_5, dim=1)
+        return seg_output
 
     def LossFN(self, x, lable):
         loss = - torch.trace(torch.mm(torch.log10(x), torch.t(lable)))