fix the bugs

fairseq/modules/speech_to_text/subsampling.py

@@ -143,7 +143,7 @@ class Conv1dSubsampling(nn.Module):
         ) for layer_id in range(num_layers)])
         ) for layer_id in range(num_layers)])
 
 
         out_dim = filters[-1]
         out_dim = filters[-1]
-        self.linear = nn.Linear(filters, filters)
+        self.linear = nn.Linear(out_dim, out_dim)
 
 
     def forward(self, x, x_len):
     def forward(self, x, x_len):