Update train.py

add num_features to input args

egs/librispeech/WSASR/conformer_ctc2/train.py

@@ -31,6 +31,7 @@ export CUDA_VISIBLE_DEVICES="0,1,2,3"
   --exp-dir conformer_ctc2/exp \
   --lang-dir data/lang_bpe_200 \
   --otc-token "<star>" \
+  --num_features 80
   --allow-bypass-arc true \
   --allow-self-loop-arc true \
   --initial-bypass-weight -19 \
@@ -160,6 +161,14 @@ def get_parser():
         """,
     )
   
+    parser.add_argument(
+        "--num_features",
+        type=int,
+        default=768,
+        help="""Number of features extracted in feature extraction stage.last dimension of feature vector.
+        80 when using fbank features and 768 or 1024 whn using wave2vec""",
+    )
+
     parser.add_argument(
         "--initial-lr",
         type=float,
@@ -373,6 +382,9 @@ def get_params() -> AttributeDict:
 
         - warm_step: The warm_step for Noam optimizer.
     """
+    parser = get_parser()
+    args = parser.parse_args()
+    feature_dim = args.num_features
     params = AttributeDict(
         {
             "best_train_loss": float("inf"),
@@ -385,7 +397,7 @@ def get_params() -> AttributeDict:
             "valid_interval": 800,  # For the 100h subset, use 800
             "alignment_interval": 25,
             # parameters for conformer
-            "feature_dim": 80, # when using fbank features for training
+            "feature_dim": feature_dim,
             "subsampling_factor": 2,
             "encoder_dim": 512,
             "nhead": 8,