AttentivePooling bugfix

nnet/xvector.py

@@ -308,11 +308,11 @@ class AttentivePooling(torch.nn.Module):
         # TODO Make convolution parameters configurable
         super(AttentivePooling, self).__init__()
         self.attention = torch.nn.Sequential(
-            torch.nn.Conv1d(num_channels * (n_mels//8), num_channels//32, kernel_size=1),
+            torch.nn.Conv1d(num_channels * (n_mels//8), num_channels//2, kernel_size=1),
             torch.nn.ReLU(),
-            torch.nn.BatchNorm1d(num_channels//32),
+            torch.nn.BatchNorm1d(num_channels//2),
             torch.nn.Tanh(),
-            torch.nn.Conv1d(num_channels//32, num_channels * (n_mels//8), kernel_size=1),
+            torch.nn.Conv1d(num_channels//2, num_channels * (n_mels//8), kernel_size=1),
             torch.nn.Softmax(dim=2),
             )
         #self.global_context = MeanStdPooling()
@@ -516,15 +516,14 @@ class Xtractor(torch.nn.Module):
             self.before_speaker_embedding_weight_decay = 0.00
             self.after_speaker_embedding_weight_decay = 0.00
         elif model_archi == "halfresnet34":
-            self.preprocessor = MelSpecFrontEnd(n_fft=512, win_length=400, hop_length=160, n_mels=64)
-            #self.preprocessor = MelSpecFrontEnd()
+            self.preprocessor = MelSpecFrontEnd()
             self.sequence_network = PreHalfResNet34()
             self.embedding_size = 512
 
-            self.before_speaker_embedding = torch.nn.Linear(in_features = 4096,
+            self.before_speaker_embedding = torch.nn.Linear(in_features = 2560,
                                                             out_features = self.embedding_size)
 
-            self.stat_pooling = AttentivePooling(256, 64)
+            self.stat_pooling = AttentivePooling(256, 80)
             self.stat_pooling_weight_decay = 0
 
             self.loss = loss