diff --git a/examples/speechcommand_v1/s0/conf/mdtc.yaml b/examples/speechcommand_v1/s0/conf/mdtc.yaml
index 6b5646f..80a9c6d 100644
--- a/examples/speechcommand_v1/s0/conf/mdtc.yaml
+++ b/examples/speechcommand_v1/s0/conf/mdtc.yaml
@@ -42,7 +42,7 @@ model:
 
 optim: adam
 optim_conf:
-    lr: 0.0002
+    lr: 0.001
     weight_decay: 0.00005
 
 training_config:
diff --git a/examples/speechcommand_v1/s0/run.sh b/examples/speechcommand_v1/s0/run.sh
index 5e58c44..27f194f 100755
--- a/examples/speechcommand_v1/s0/run.sh
+++ b/examples/speechcommand_v1/s0/run.sh
@@ -6,7 +6,7 @@
 
 export CUDA_VISIBLE_DEVICES="0"
 
-stage=1
+stage=2
 stop_stage=4
 num_keywords=11
 
diff --git a/kws/bin/train.py b/kws/bin/train.py
index 51e2766..a8cd93d 100644
--- a/kws/bin/train.py
+++ b/kws/bin/train.py
@@ -196,7 +196,8 @@ def main():
         model = model.to(device)
 
     optimizer = optim.Adam(model.parameters(),
-                           lr=configs['optim_conf']['lr'])
+                           lr=configs['optim_conf']['lr'],
+                           weight_decay=configs['optim_conf']['weight_decay'])
     scheduler = optim.lr_scheduler.ReduceLROnPlateau(
         optimizer,
         mode='min',