diff --git a/paddlespeech/s2t/exps/u2/model.py b/paddlespeech/s2t/exps/u2/model.py
index b6dbcf44..5dbb72f4 100644
--- a/paddlespeech/s2t/exps/u2/model.py
+++ b/paddlespeech/s2t/exps/u2/model.py
@@ -128,8 +128,9 @@ class U2Trainer(Trainer):
             if dist.get_rank() == 0 and self.visualizer:
                 losses_np_v = losses_np.copy()
                 losses_np_v.update({"lr": self.lr_scheduler()})
-                self.visualizer.add_scalars("step", losses_np_v,
-                                            self.iteration - 1)
+                for key, val in losses_np_v.items():
+                    self.visualizer.add_scalar(tag='train/'+key, value=val, step=self.iteration-1)
+
 
     @paddle.no_grad()
     def valid(self):
@@ -237,9 +238,8 @@ class U2Trainer(Trainer):
             logger.info(
                 'Epoch {} Val info val_loss {}'.format(self.epoch, cv_loss))
             if self.visualizer:
-                self.visualizer.add_scalars(
-                    'epoch', {'cv_loss': cv_loss,
-                              'lr': self.lr_scheduler()}, self.epoch)
+                self.visualizer.add_scalar(tag='eval/cv_loss', value=cv_loss, step=self.epoch)
+                self.visualizer.add_scalar(tag='eval/lr', value=self.lr_scheduler(), step=self.epoch)
 
             self.save(tag=self.epoch, infos={'val_loss': cv_loss})
             self.new_epoch()
diff --git a/paddlespeech/s2t/exps/u2_kaldi/model.py b/paddlespeech/s2t/exps/u2_kaldi/model.py
index c23b4c24..a3f45d8e 100644
--- a/paddlespeech/s2t/exps/u2_kaldi/model.py
+++ b/paddlespeech/s2t/exps/u2_kaldi/model.py
@@ -131,8 +131,8 @@ class U2Trainer(Trainer):
             if dist.get_rank() == 0 and self.visualizer:
                 losses_np_v = losses_np.copy()
                 losses_np_v.update({"lr": self.lr_scheduler()})
-                self.visualizer.add_scalars("step", losses_np_v,
-                                            self.iteration - 1)
+                for key, val in losses_np_v.items():
+                    self.visualizer.add_scalar(tag="train/"+key, value=val, step=self.iteration - 1)
 
     @paddle.no_grad()
     def valid(self):
@@ -222,9 +222,9 @@ class U2Trainer(Trainer):
             logger.info(
                 'Epoch {} Val info val_loss {}'.format(self.epoch, cv_loss))
             if self.visualizer:
-                self.visualizer.add_scalars(
-                    'epoch', {'cv_loss': cv_loss,
-                              'lr': self.lr_scheduler()}, self.epoch)
+                self.visualizer.add_scalar(tag='eval/cv_loss', value=cv_loss, step=self.epoch)
+                self.visualizer.add_scalar(tag='eval/lr', value=self.lr_scheduler(), step=self.epoch)
+                
             self.save(tag=self.epoch, infos={'val_loss': cv_loss})
             self.new_epoch()
 
diff --git a/paddlespeech/s2t/exps/u2_st/model.py b/paddlespeech/s2t/exps/u2_st/model.py
index 034463fe..771203cf 100644
--- a/paddlespeech/s2t/exps/u2_st/model.py
+++ b/paddlespeech/s2t/exps/u2_st/model.py
@@ -138,8 +138,8 @@ class U2STTrainer(Trainer):
             if dist.get_rank() == 0 and self.visualizer:
                 losses_np_v = losses_np.copy()
                 losses_np_v.update({"lr": self.lr_scheduler()})
-                self.visualizer.add_scalars("step", losses_np_v,
-                                            self.iteration - 1)
+                for key, val in losses_np_v.items():
+                    self.visualizer.add_scalar(tag="train/"+key, value=val, step=self.iteration - 1)
 
     @paddle.no_grad()
     def valid(self):
@@ -235,9 +235,9 @@ class U2STTrainer(Trainer):
             logger.info(
                 'Epoch {} Val info val_loss {}'.format(self.epoch, cv_loss))
             if self.visualizer:
-                self.visualizer.add_scalars(
-                    'epoch', {'cv_loss': cv_loss,
-                              'lr': self.lr_scheduler()}, self.epoch)
+                self.visualizer.add_scalar(tag='eval/cv_loss', value=cv_loss, step=self.epoch)
+                self.visualizer.add_scalar(tag='eval/lr', value=self.lr_scheduler(), step=self.epoch)
+                
             self.save(tag=self.epoch, infos={'val_loss': cv_loss})
             self.new_epoch()
 
diff --git a/paddlespeech/s2t/frontend/normalizer.py b/paddlespeech/s2t/frontend/normalizer.py
index a29cddc3..c55ec9a3 100644
--- a/paddlespeech/s2t/frontend/normalizer.py
+++ b/paddlespeech/s2t/frontend/normalizer.py
@@ -16,19 +16,36 @@ import json
 
 import numpy as np
 import paddle
+import jsonlines
 from paddle.io import DataLoader
 from paddle.io import Dataset
 
 from paddlespeech.s2t.frontend.audio import AudioSegment
 from paddlespeech.s2t.frontend.utility import load_cmvn
-from paddlespeech.s2t.frontend.utility import read_manifest
 from paddlespeech.s2t.utils.log import Log
 
 __all__ = ["FeatureNormalizer"]
 
 logger = Log(__name__).getlog()
 
-
+def read_manifest(manifest_path):
+     """Load and parse manifest file.
+ 
+     Args:
+         manifest_path ([type]): Manifest file to load and parse.
+     Raises:
+         IOError: If failed to parse the manifest.
+ 
+     Returns:
+         List[dict]: Manifest parsing results.
+     """
+ 
+     manifest = []
+     with jsonlines.open(manifest_path, 'r') as reader:
+         for json_data in reader:
+            manifest.append(json_data)
+     return manifest
+ 
 # https://github.com/PaddlePaddle/Paddle/pull/31481
 class CollateFunc(object):
     def __init__(self, feature_func):
@@ -61,7 +78,11 @@ class CollateFunc(object):
 class AudioDataset(Dataset):
     def __init__(self, manifest_path, num_samples=-1, rng=None, random_seed=0):
         self._rng = rng if rng else np.random.RandomState(random_seed)
-        manifest = read_manifest(manifest_path)
+        manifest = []
+        with jsonlines.open(manifest_path, 'r') as reader:
+         for json_data in reader:
+            manifest.append(json_data)
+        
         if num_samples == -1:
             sampled_manifest = manifest
         else:
diff --git a/paddlespeech/s2t/frontend/utility.py b/paddlespeech/s2t/frontend/utility.py
index d423a604..948aba06 100644
--- a/paddlespeech/s2t/frontend/utility.py
+++ b/paddlespeech/s2t/frontend/utility.py
@@ -65,7 +65,26 @@ def load_dict(dict_path: Optional[Text], maskctc=False) -> Optional[List[Text]]:
     return char_list
 
 
-def read_manifest(
+def read_manifest(manifest_path,):
+    """Load and parse manifest file.
+
+    Args:
+        manifest_path ([type]): Manifest file to load and parse.
+
+    Raises:
+        IOError: If failed to parse the manifest.
+
+    Returns:
+        List[dict]: Manifest parsing results.
+    """
+    manifest = []
+    with jsonlines.open(manifest_path, 'r') as reader:
+        for json_data in reader:
+            manifest.append(json_data)
+    return manifest
+
+    
+def read_manifest_filter(
         manifest_path,
         max_input_len=float('inf'),
         min_input_len=0.0,
@@ -98,7 +117,6 @@ def read_manifest(
     Returns:
         List[dict]: Manifest parsing results.
     """
-
     manifest = []
     with jsonlines.open(manifest_path, 'r') as reader:
         for json_data in reader:
diff --git a/paddlespeech/s2t/io/dataset.py b/paddlespeech/s2t/io/dataset.py
index 61eeb00f..006cfe04 100644
--- a/paddlespeech/s2t/io/dataset.py
+++ b/paddlespeech/s2t/io/dataset.py
@@ -95,7 +95,7 @@ class ManifestDataset(Dataset):
         super().__init__()
 
         # read manifest
-        self._manifest = read_manifest(
+        self._manifest = read_manifest_filter(
             manifest_path=manifest_path,
             max_input_len=max_input_len,
             min_input_len=min_input_len,
diff --git a/paddlespeech/s2t/training/trainer.py b/paddlespeech/s2t/training/trainer.py
index f5fb2db0..be398814 100644
--- a/paddlespeech/s2t/training/trainer.py
+++ b/paddlespeech/s2t/training/trainer.py
@@ -19,7 +19,7 @@ from pathlib import Path
 
 import paddle
 from paddle import distributed as dist
-from tensorboardX import SummaryWriter
+from visualdl import LogWriter
 
 from paddlespeech.s2t.training.reporter import ObsScope
 from paddlespeech.s2t.training.reporter import report
@@ -309,9 +309,8 @@ class Trainer():
             logger.info(
                 'Epoch {} Val info val_loss {}'.format(self.epoch, cv_loss))
             if self.visualizer:
-                self.visualizer.add_scalars(
-                    'epoch', {'cv_loss': cv_loss,
-                              'lr': self.lr_scheduler()}, self.epoch)
+                self.visualizer.add_scalar(tag='eval/cv_loss', value=cv_loss, step=self.epoch)
+                self.visualizer.add_scalar(tag='eval/lr', value=self.lr_scheduler(), step=self.epoch)
 
             # after epoch
             self.save(tag=self.epoch, infos={'val_loss': cv_loss})
@@ -427,7 +426,7 @@ class Trainer():
         unexpected behaviors.
         """
         # visualizer
-        visualizer = SummaryWriter(logdir=str(self.visual_dir))
+        visualizer = LogWriter(logdir=str(self.visual_dir))
         self.visualizer = visualizer
 
     @mp_tools.rank_zero_only
diff --git a/paddlespeech/text/speechtask/punctuation_restoration/training/trainer.py b/paddlespeech/text/speechtask/punctuation_restoration/training/trainer.py
index d6b6eeb6..ba7ddde3 100644
--- a/paddlespeech/text/speechtask/punctuation_restoration/training/trainer.py
+++ b/paddlespeech/text/speechtask/punctuation_restoration/training/trainer.py
@@ -34,7 +34,7 @@ from speechtask.punctuation_restoration.model.lstm import RnnLm
 from speechtask.punctuation_restoration.utils import layer_tools
 from speechtask.punctuation_restoration.utils import mp_tools
 from speechtask.punctuation_restoration.utils.checkpoint import Checkpoint
-from tensorboardX import SummaryWriter
+from visualdl import LogWriter
 
 __all__ = ["Trainer", "Tester"]
 
@@ -252,10 +252,8 @@ class Trainer():
             self.logger.info("Epoch {} Val info val_loss {}, F1_score {}".
                              format(self.epoch, total_loss, F1_score))
             if self.visualizer:
-                self.visualizer.add_scalars("epoch", {
-                    "total_loss": total_loss,
-                    "lr": self.lr_scheduler()
-                }, self.epoch)
+                self.visualizer.add_scalar(tag='eval/cv_loss', value=cv_loss, step=self.epoch)
+                self.visualizer.add_scalar(tag='eval/lr', value=self.lr_scheduler(), step=self.epoch)
 
             self.save(
                 tag=self.epoch, infos={"val_loss": total_loss,
@@ -341,7 +339,7 @@ class Trainer():
         unexpected behaviors.
         """
         # visualizer
-        visualizer = SummaryWriter(logdir=str(self.output_dir))
+        visualizer = LogWriter(logdir=str(self.output_dir))
         self.visualizer = visualizer
 
     @mp_tools.rank_zero_only
diff --git a/requirements.txt b/requirements.txt
index 99e485f8..2ee60d3f 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -40,7 +40,6 @@ snakeviz
 soundfile~=0.10
 sox
 soxbindings
-tensorboardX
 textgrid
 timer
 tqdm
diff --git a/utils/build_vocab.py b/utils/build_vocab.py
index 6a903147..61dc5e25 100755
--- a/utils/build_vocab.py
+++ b/utils/build_vocab.py
@@ -19,11 +19,11 @@ import argparse
 import functools
 import os
 import tempfile
+import jsonlines
 from collections import Counter
 
 from paddlespeech.s2t.frontend.featurizer.text_featurizer import TextFeaturizer
 from paddlespeech.s2t.frontend.utility import BLANK
-from paddlespeech.s2t.frontend.utility import read_manifest
 from paddlespeech.s2t.frontend.utility import SOS
 from paddlespeech.s2t.frontend.utility import SPACE
 from paddlespeech.s2t.frontend.utility import UNK
@@ -59,13 +59,21 @@ args = parser.parse_args()
 
 
 def count_manifest(counter, text_feature, manifest_path):
-    manifest_jsons = read_manifest(manifest_path)
+    manifest_jsons = []
+    with jsonlines.open(manifest_path, 'r') as reader:
+        for json_data in reader:
+            manifest_jsons.append(json_data)
+        
     for line_json in manifest_jsons:
         line = text_feature.tokenize(line_json['text'], replace_space=False)
         counter.update(line)
 
 def dump_text_manifest(fileobj, manifest_path, key='text'):
-    manifest_jsons = read_manifest(manifest_path)
+    manifest_jsons = []
+    with jsonlines.open(manifest_path, 'r') as reader:
+        for json_data in reader:
+            manifest_jsons.append(json_data)
+            
     for line_json in manifest_jsons:
         fileobj.write(line_json[key] + "\n")
 
diff --git a/utils/utility.py b/utils/utility.py
index b4db518a..29fda268 100755
--- a/utils/utility.py
+++ b/utils/utility.py
@@ -42,6 +42,7 @@ def read_manifest(manifest_path):
     for json_line in open(manifest_path, 'r'):
         try:
             json_data = json.loads(json_line)
+            manifest.append(json_data)
         except Exception as e:
             raise IOError("Error reading manifest: %s" % str(e))
     return manifest