Merge pull request #2 from wd929/icmc

Decoding scripts

egs/icmcasr/ASR/zipformer/decode.py

@@ -2,6 +2,7 @@
 #
 # Copyright 2021-2023 Xiaomi Corporation (Author: Fangjun Kuang,
 #                                                 Zengwei Yao)
+#                2023 NVIDIA Corporation (Wen Ding)
 #
 # See ../../../../LICENSE for clarification regarding multiple authors
 #
@@ -20,9 +21,9 @@
 Usage:
 (1) greedy search
 ./zipformer/decode.py \
-    --epoch 28 \
-    --avg 15 \
-    --exp-dir ./zipformer/exp \
+    --epoch 13 \
+    --avg 10 \
+    --exp-dir ./zipformer/exp_maxdur500_lr0.0225/ \
     --max-duration 600 \
     --decoding-method greedy_search
 
@@ -106,7 +107,7 @@ import k2
 import sentencepiece as spm
 import torch
 import torch.nn as nn
-from asr_datamodule import LibriSpeechAsrDataModule
+from asr_datamodule import ICMCAsrDataModule
 from beam_search import (
     beam_search,
     fast_beam_search_nbest,
@@ -729,7 +730,7 @@ def save_results(
         )
         with open(errs_filename, "w") as f:
             wer = write_error_stats(
-                f, f"{test_set_name}-{key}", results, enable_log=True
+                f, f"{test_set_name}-{key}", results, enable_log=True, compute_CER=True
             )
             test_set_wers[key] = wer
 
@@ -755,7 +756,7 @@ def save_results(
 @torch.no_grad()
 def main():
     parser = get_parser()
-    LibriSpeechAsrDataModule.add_arguments(parser)
+    ICMCAsrDataModule.add_arguments(parser)
     LmScorer.add_arguments(parser)
     args = parser.parse_args()
     args.exp_dir = Path(args.exp_dir)
@@ -1014,16 +1015,16 @@ def main():
 
     # we need cut ids to display recognition results.
     args.return_cuts = True
-    librispeech = LibriSpeechAsrDataModule(args)
+    icmc = ICMCAsrDataModule(args)
 
-    test_clean_cuts = librispeech.test_clean_cuts()
-    test_other_cuts = librispeech.test_other_cuts()
+    test_ihm_cuts = icmc.dev_ihm_cuts()
+    test_shm_cuts = icmc.dev_shm_cuts()
 
-    test_clean_dl = librispeech.test_dataloaders(test_clean_cuts)
-    test_other_dl = librispeech.test_dataloaders(test_other_cuts)
+    test_ihm_dl = icmc.test_dataloaders(test_ihm_cuts)
+    test_shm_dl = icmc.test_dataloaders(test_shm_cuts)
 
-    test_sets = ["test-clean", "test-other"]
-    test_dl = [test_clean_dl, test_other_dl]
+    test_sets = ["test-ihm", "test-shm"]
+    test_dl = [test_ihm_dl, test_shm_dl]
 
     for test_set, test_dl in zip(test_sets, test_dl):
         results_dict = decode_dataset(