update ctc-decoding for pretrained.py on conformer_ctc

docs/source/recipes/librispeech/conformer_ctc.rst

@@ -448,7 +448,7 @@ After downloading, you will have the following files:
 **File descriptions**:
   - ``data/lang_bpe/Linv.pt``
 
-      It is the lexicon file.
+      It is the lexicon file, with word IDs as labels and token IDs as aux_labels.
 
   - ``data/lang_bpe/HLG.pt``
 
@@ -530,7 +530,7 @@ Usage
 
 displays the help information.
 
-It supports three decoding methods:
+It supports 4 decoding methods:
 
   - CTC decoding
   - HLG decoding

egs/librispeech/ASR/conformer_ctc/pretrained.py

@@ -57,16 +57,14 @@ def get_parser():
     parser.add_argument(
         "--words-file",
         type=str,
-        default="./tmp/icefall_asr_librispeech_conformer_ctc/ \
-        data/lang_bpe/words.txt",
+        required=True,
         help="Path to words.txt",
     )
 
     parser.add_argument(
         "--HLG",
         type=str,
-        default="./tmp/icefall_asr_librispeech_conformer_ctc/ \
-        data/lang_bpe/HLG.pt",
+        required=True,
         help="Path to HLG.pt.",
     )
 
@@ -172,8 +170,7 @@ def get_parser():
     parser.add_argument(
         "--lang-dir",
         type=str,
-        default="./tmp/icefall_asr_librispeech_conformer_ctc/ \
-        data/lang_bpe",
+        required=True,
         help="Path to lang bpe dir.",
     )
 
@@ -302,6 +299,7 @@ def main():
         dtype=torch.int32,
     )
 
+    try:
         if params.method == "ctc-decoding":
             logging.info("Building CTC topology")
             lexicon = Lexicon(params.lang_dir)
@@ -335,7 +333,11 @@ def main():
             hyps = bpe_model.decode(token_ids)
             hyps = [s.split() for s in hyps]
 
-    else:
+        if params.method in [
+            "1best",
+            "whole-lattice-rescoring",
+            "attention-decoder",
+        ]:
             logging.info(f"Loading HLG from {params.HLG}")
             HLG = k2.Fsa.from_dict(torch.load(params.HLG, map_location="cpu"))
             HLG = HLG.to(device)
@@ -343,7 +345,10 @@ def main():
                 # For whole-lattice-rescoring and attention-decoder
                 HLG.lm_scores = HLG.scores.clone()
 
-        if params.method in ["whole-lattice-rescoring", "attention-decoder"]:
+            if params.method in [
+                "whole-lattice-rescoring",
+                "attention-decoder",
+            ]:
                 logging.info(f"Loading G from {params.G}")
                 G = k2.Fsa.from_dict(torch.load(params.G, map_location="cpu"))
                 # Add epsilon self-loops to G as we will compose
@@ -378,7 +383,9 @@ def main():
                 )
                 best_path = next(iter(best_path_dict.values()))
             elif params.method == "attention-decoder":
-            logging.info("Use HLG + LM rescoring + attention decoder rescoring")
+                logging.info(
+                    "Use HLG + LM rescoring + attention decoder rescoring"
+                )
                 rescored_lattice = rescore_with_whole_lattice(
                     lattice=lattice, G_with_epsilon_loops=G, lm_scale_list=None
                 )
@@ -408,6 +415,9 @@ def main():
 
         logging.info("Decoding Done")
 
+    except Exception:
+        raise ValueError("Please use a supported decoding method.")
+
 
 if __name__ == "__main__":
     formatter = (