diff --git a/egs/aidatatang_200zh/ASR/local/compute_fbank_aidatatang_200zh.py b/egs/aidatatang_200zh/ASR/local/compute_fbank_aidatatang_200zh.py
index 0b54fcb9a..fb2751c0f 100755
--- a/egs/aidatatang_200zh/ASR/local/compute_fbank_aidatatang_200zh.py
+++ b/egs/aidatatang_200zh/ASR/local/compute_fbank_aidatatang_200zh.py
@@ -62,6 +62,13 @@ def compute_fbank_aidatatang_200zh(num_mel_bins: int = 80):
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
     with get_executor() as ex:  # Initialize the executor only once.
diff --git a/egs/aishell/ASR/local/compute_fbank_aidatatang_200zh.py b/egs/aishell/ASR/local/compute_fbank_aidatatang_200zh.py
index 8cdfad71f..42700a972 100755
--- a/egs/aishell/ASR/local/compute_fbank_aidatatang_200zh.py
+++ b/egs/aishell/ASR/local/compute_fbank_aidatatang_200zh.py
@@ -62,6 +62,13 @@ def compute_fbank_aidatatang_200zh(num_mel_bins: int = 80):
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
     with get_executor() as ex:  # Initialize the executor only once.
diff --git a/egs/aishell/ASR/local/compute_fbank_aishell.py b/egs/aishell/ASR/local/compute_fbank_aishell.py
index e27e35ec5..deab6c809 100755
--- a/egs/aishell/ASR/local/compute_fbank_aishell.py
+++ b/egs/aishell/ASR/local/compute_fbank_aishell.py
@@ -62,6 +62,13 @@ def compute_fbank_aishell(num_mel_bins: int = 80):
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
     with get_executor() as ex:  # Initialize the executor only once.
diff --git a/egs/aishell2/ASR/local/compute_fbank_aishell2.py b/egs/aishell2/ASR/local/compute_fbank_aishell2.py
index 7bc969a1a..d8d3622bd 100755
--- a/egs/aishell2/ASR/local/compute_fbank_aishell2.py
+++ b/egs/aishell2/ASR/local/compute_fbank_aishell2.py
@@ -62,6 +62,13 @@ def compute_fbank_aishell2(num_mel_bins: int = 80):
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
     with get_executor() as ex:  # Initialize the executor only once.
diff --git a/egs/aishell4/ASR/local/compute_fbank_aishell4.py b/egs/aishell4/ASR/local/compute_fbank_aishell4.py
index 09f885636..3f50d9e3e 100755
--- a/egs/aishell4/ASR/local/compute_fbank_aishell4.py
+++ b/egs/aishell4/ASR/local/compute_fbank_aishell4.py
@@ -63,6 +63,13 @@ def compute_fbank_aishell4(num_mel_bins: int = 80):
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
     with get_executor() as ex:  # Initialize the executor only once.
diff --git a/egs/alimeeting/ASR/local/compute_fbank_alimeeting.py b/egs/alimeeting/ASR/local/compute_fbank_alimeeting.py
index eb3acbbea..af926aa53 100755
--- a/egs/alimeeting/ASR/local/compute_fbank_alimeeting.py
+++ b/egs/alimeeting/ASR/local/compute_fbank_alimeeting.py
@@ -63,6 +63,13 @@ def compute_fbank_alimeeting(num_mel_bins: int = 80):
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
     with get_executor() as ex:  # Initialize the executor only once.
diff --git a/egs/gigaspeech/ASR/local/preprocess_gigaspeech.py b/egs/gigaspeech/ASR/local/preprocess_gigaspeech.py
index 0cec82ad5..48d10a157 100755
--- a/egs/gigaspeech/ASR/local/preprocess_gigaspeech.py
+++ b/egs/gigaspeech/ASR/local/preprocess_gigaspeech.py
@@ -62,6 +62,13 @@ def preprocess_giga_speech():
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     for partition, m in manifests.items():
         logging.info(f"Processing {partition}")
         raw_cuts_path = output_dir / f"cuts_{partition}_raw.jsonl.gz"
diff --git a/egs/librispeech/ASR/local/compute_fbank_librispeech.py b/egs/librispeech/ASR/local/compute_fbank_librispeech.py
index 642d9fd32..f3e15e039 100755
--- a/egs/librispeech/ASR/local/compute_fbank_librispeech.py
+++ b/egs/librispeech/ASR/local/compute_fbank_librispeech.py
@@ -66,6 +66,13 @@ def compute_fbank_librispeech():
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
     with get_executor() as ex:  # Initialize the executor only once.
diff --git a/egs/librispeech/ASR/local/compute_fbank_musan.py b/egs/librispeech/ASR/local/compute_fbank_musan.py
index fef372129..056da29e5 100755
--- a/egs/librispeech/ASR/local/compute_fbank_musan.py
+++ b/egs/librispeech/ASR/local/compute_fbank_musan.py
@@ -65,6 +65,8 @@ def compute_fbank_musan():
     assert len(manifests) == len(dataset_parts), (
         len(manifests),
         len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
     )
 
     musan_cuts_path = output_dir / "musan_cuts.jsonl.gz"
diff --git a/egs/librispeech/ASR/local/preprocess_gigaspeech.py b/egs/librispeech/ASR/local/preprocess_gigaspeech.py
index 0f4ae820b..077f23039 100644
--- a/egs/librispeech/ASR/local/preprocess_gigaspeech.py
+++ b/egs/librispeech/ASR/local/preprocess_gigaspeech.py
@@ -68,6 +68,13 @@ def preprocess_giga_speech():
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     for partition, m in manifests.items():
         logging.info(f"Processing {partition}")
         raw_cuts_path = output_dir / f"{prefix}_cuts_{partition}_raw.{suffix}"
diff --git a/egs/librispeech/ASR/pruned_transducer_stateless3/test_scaling_converter.py b/egs/librispeech/ASR/pruned_transducer_stateless3/test_scaling_converter.py
index a9feea83c..2e131158f 100644
--- a/egs/librispeech/ASR/pruned_transducer_stateless3/test_scaling_converter.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless3/test_scaling_converter.py
@@ -181,7 +181,7 @@ def test_convert_scaled_to_non_scaled():
         y = torch.randint(low=1, high=vocab_size - 1, size=(N, U))
 
         d1 = model.decoder(y)
-        d2 = model.decoder(y)
+        d2 = converted_model.decoder(y)
 
         assert torch.allclose(d1, d2)
 
diff --git a/egs/spgispeech/ASR/local/compute_fbank_musan.py b/egs/spgispeech/ASR/local/compute_fbank_musan.py
index 70372af2b..6cb8b65ae 100755
--- a/egs/spgispeech/ASR/local/compute_fbank_musan.py
+++ b/egs/spgispeech/ASR/local/compute_fbank_musan.py
@@ -69,6 +69,13 @@ def compute_fbank_musan():
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     musan_cuts_path = src_dir / "cuts_musan.jsonl.gz"
 
     if musan_cuts_path.is_file():
diff --git a/egs/tal_csasr/ASR/local/compute_fbank_tal_csasr.py b/egs/tal_csasr/ASR/local/compute_fbank_tal_csasr.py
index 367e098f7..4582609ac 100755
--- a/egs/tal_csasr/ASR/local/compute_fbank_tal_csasr.py
+++ b/egs/tal_csasr/ASR/local/compute_fbank_tal_csasr.py
@@ -62,6 +62,13 @@ def compute_fbank_tal_csasr(num_mel_bins: int = 80):
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
     with get_executor() as ex:  # Initialize the executor only once.
diff --git a/egs/tedlium3/ASR/local/compute_fbank_tedlium.py b/egs/tedlium3/ASR/local/compute_fbank_tedlium.py
index e324b5025..327962a79 100755
--- a/egs/tedlium3/ASR/local/compute_fbank_tedlium.py
+++ b/egs/tedlium3/ASR/local/compute_fbank_tedlium.py
@@ -62,6 +62,13 @@ def compute_fbank_tedlium():
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
     with get_executor() as ex:  # Initialize the executor only once.
diff --git a/egs/timit/ASR/local/compute_fbank_timit.py b/egs/timit/ASR/local/compute_fbank_timit.py
index 094769c8c..f25786a0c 100644
--- a/egs/timit/ASR/local/compute_fbank_timit.py
+++ b/egs/timit/ASR/local/compute_fbank_timit.py
@@ -63,6 +63,13 @@ def compute_fbank_timit():
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(FbankConfig(num_mel_bins=num_mel_bins))
 
     with get_executor() as ex:  # Initialize the executor only once.
diff --git a/egs/wenetspeech/ASR/local/preprocess_wenetspeech.py b/egs/wenetspeech/ASR/local/preprocess_wenetspeech.py
index 64733eb15..817969c47 100755
--- a/egs/wenetspeech/ASR/local/preprocess_wenetspeech.py
+++ b/egs/wenetspeech/ASR/local/preprocess_wenetspeech.py
@@ -23,6 +23,8 @@ from pathlib import Path
 from lhotse import CutSet, SupervisionSegment
 from lhotse.recipes.utils import read_manifests_if_cached
 
+from icefall import setup_logger
+
 # Similar text filtering and normalization procedure as in:
 # https://github.com/SpeechColab/WenetSpeech/blob/main/toolkits/kaldi/wenetspeech_data_prep.sh
 
@@ -48,13 +50,17 @@ def preprocess_wenet_speech():
     output_dir = Path("data/fbank")
     output_dir.mkdir(exist_ok=True)
 
+    # Note: By default, we preprocess all sub-parts.
+    # You can delete those that you don't need.
+    # For instance, if you don't want to use the L subpart, just remove
+    # the line below containing "L"
     dataset_parts = (
-        "L",
-        "M",
-        "S",
         "DEV",
         "TEST_NET",
         "TEST_MEETING",
+        "S",
+        "M",
+        "L",
     )
 
     logging.info("Loading manifest (may take 10 minutes)")
@@ -66,6 +72,13 @@ def preprocess_wenet_speech():
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     for partition, m in manifests.items():
         logging.info(f"Processing {partition}")
         raw_cuts_path = output_dir / f"cuts_{partition}_raw.jsonl.gz"
@@ -81,10 +94,13 @@ def preprocess_wenet_speech():
         logging.info(f"Normalizing text in {partition}")
         for sup in m["supervisions"]:
             text = str(sup.text)
-            logging.info(f"Original text: {text}")
+            orig_text = text
             sup.text = normalize_text(sup.text)
             text = str(sup.text)
-            logging.info(f"Normalize text: {text}")
+            if len(orig_text) != len(text):
+                logging.info(
+                    f"\nOriginal text vs normalized text:\n{orig_text}\n{text}"
+                )
 
         # Create long-recording cut manifests.
         logging.info(f"Processing {partition}")
@@ -109,12 +125,10 @@ def preprocess_wenet_speech():
 
 
 def main():
-    formatter = (
-        "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
-    )
-    logging.basicConfig(format=formatter, level=logging.INFO)
+    setup_logger(log_filename="./log-preprocess-wenetspeech")
 
     preprocess_wenet_speech()
+    logging.info("Done")
 
 
 if __name__ == "__main__":
diff --git a/egs/wenetspeech/ASR/pruned_transducer_stateless2/train.py b/egs/wenetspeech/ASR/pruned_transducer_stateless2/train.py
index 5208dbefe..d3cc7c9c9 100644
--- a/egs/wenetspeech/ASR/pruned_transducer_stateless2/train.py
+++ b/egs/wenetspeech/ASR/pruned_transducer_stateless2/train.py
@@ -81,7 +81,6 @@ For training with the S subset:
 
 import argparse
 import logging
-import os
 import warnings
 from pathlib import Path
 from shutil import copyfile
@@ -120,8 +119,6 @@ LRSchedulerType = Union[
     torch.optim.lr_scheduler._LRScheduler, optim.LRScheduler
 ]
 
-os.environ["CUDA_LAUNCH_BLOCKING"] = "1"
-
 
 def get_parser():
     parser = argparse.ArgumentParser(
@@ -162,7 +159,7 @@ def get_parser():
         default=0,
         help="""Resume training from from this epoch.
         If it is positive, it will load checkpoint from
-        transducer_stateless2/exp/epoch-{start_epoch-1}.pt
+        pruned_transducer_stateless2/exp/epoch-{start_epoch-1}.pt
         """,
     )
 
@@ -361,8 +358,8 @@ def get_params() -> AttributeDict:
             "best_valid_loss": float("inf"),
             "best_train_epoch": -1,
             "best_valid_epoch": -1,
-            "batch_idx_train": 10,
-            "log_interval": 1,
+            "batch_idx_train": 0,
+            "log_interval": 50,
             "reset_interval": 200,
             # parameters for conformer
             "feature_dim": 80,
@@ -545,7 +542,7 @@ def compute_loss(
     warmup: float = 1.0,
 ) -> Tuple[Tensor, MetricsTracker]:
     """
-    Compute CTC loss given the model and its inputs.
+    Compute RNN-T loss given the model and its inputs.
     Args:
       params:
         Parameters for training. See :func:`get_params`.
@@ -573,7 +570,7 @@ def compute_loss(
     texts = batch["supervisions"]["text"]
 
     y = graph_compiler.texts_to_ids(texts)
-    if type(y) == list:
+    if isinstance(y, list):
         y = k2.RaggedTensor(y).to(device)
     else:
         y = y.to(device)
@@ -697,7 +694,6 @@ def train_one_epoch(
     tot_loss = MetricsTracker()
 
     for batch_idx, batch in enumerate(train_dl):
-
         params.batch_idx_train += 1
         batch_size = len(batch["supervisions"]["text"])
 
diff --git a/egs/wenetspeech/ASR/pruned_transducer_stateless5/train.py b/egs/wenetspeech/ASR/pruned_transducer_stateless5/train.py
index 5a5925d55..2052e9da7 100755
--- a/egs/wenetspeech/ASR/pruned_transducer_stateless5/train.py
+++ b/egs/wenetspeech/ASR/pruned_transducer_stateless5/train.py
@@ -61,7 +61,6 @@ export CUDA_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
 import argparse
 import copy
 import logging
-import os
 import warnings
 from pathlib import Path
 from shutil import copyfile
@@ -103,8 +102,6 @@ LRSchedulerType = Union[
     torch.optim.lr_scheduler._LRScheduler, optim.LRScheduler
 ]
 
-os.environ["CUDA_LAUNCH_BLOCKING"] = "1"
-
 
 def add_model_arguments(parser: argparse.ArgumentParser):
     parser.add_argument(
@@ -684,7 +681,7 @@ def compute_loss(
     texts = batch["supervisions"]["text"]
 
     y = graph_compiler.texts_to_ids(texts)
-    if type(y) == list:
+    if isinstance(y, list):
         y = k2.RaggedTensor(y).to(device)
     else:
         y = y.to(device)
diff --git a/egs/yesno/ASR/local/compute_fbank_yesno.py b/egs/yesno/ASR/local/compute_fbank_yesno.py
index fb48b6f8e..9a4e8a36f 100755
--- a/egs/yesno/ASR/local/compute_fbank_yesno.py
+++ b/egs/yesno/ASR/local/compute_fbank_yesno.py
@@ -47,6 +47,13 @@ def compute_fbank_yesno():
     )
     assert manifests is not None
 
+    assert len(manifests) == len(dataset_parts), (
+        len(manifests),
+        len(dataset_parts),
+        list(manifests.keys()),
+        dataset_parts,
+    )
+
     extractor = Fbank(
         FbankConfig(sampling_rate=8000, num_mel_bins=num_mel_bins)
     )