diff --git a/docs/source/decoding-with-langugage-models/LODR.rst b/docs/source/decoding-with-langugage-models/LODR.rst
index 8cc1a624cf..b6b6e8cbb2 100644
--- a/docs/source/decoding-with-langugage-models/LODR.rst
+++ b/docs/source/decoding-with-langugage-models/LODR.rst
@@ -56,7 +56,7 @@ during decoding for transducer model:
     \lambda_1 \log p_{\text{Target LM}}\left(y_u|\mathit{x},y_{1:u-1}\right) -
     \lambda_2 \log p_{\text{bi-gram}}\left(y_u|\mathit{x},y_{1:u-1}\right)
 
-In LODR, an additional bi-gram LM estimated on the source domain (e.g training corpus) is required. Comared to DR,
+In LODR, an additional bi-gram LM estimated on the source domain (e.g training corpus) is required. Compared to DR,
 the only difference lies in the choice of source domain LM. According to the original `paper <https://arxiv.org/abs/2203.16776>`_,
 LODR achieves similar performance compared DR in both intra-domain and cross-domain settings.
 As a bi-gram is much faster to evaluate, LODR is usually much faster.
diff --git a/docs/source/model-export/export-ncnn-conv-emformer.rst b/docs/source/model-export/export-ncnn-conv-emformer.rst
index 4f5535d831..93392aee7a 100644
--- a/docs/source/model-export/export-ncnn-conv-emformer.rst
+++ b/docs/source/model-export/export-ncnn-conv-emformer.rst
@@ -125,7 +125,7 @@ Python code. We have also set up ``PATH`` so that you can use
 .. caution::
 
   Please don't use `<https://github.com/tencent/ncnn>`_.
-  We have made some modifications to the offical `ncnn`_.
+  We have made some modifications to the official `ncnn`_.
 
   We will synchronize `<https://github.com/csukuangfj/ncnn>`_ periodically
   with the official one.
diff --git a/egs/wenetspeech/ASR/pruned_transducer_stateless2/decode.py b/egs/wenetspeech/ASR/pruned_transducer_stateless2/decode.py
index bdd1f27bc6..2bafe25d6b 100755
--- a/egs/wenetspeech/ASR/pruned_transducer_stateless2/decode.py
+++ b/egs/wenetspeech/ASR/pruned_transducer_stateless2/decode.py
@@ -203,7 +203,7 @@ def get_parser():
         "--beam-size",
         type=int,
         default=4,
-        help="""An interger indicating how many candidates we will keep for each
+        help="""An integer indicating how many candidates we will keep for each
         frame. Used only when --decoding-method is beam_search or
         modified_beam_search.""",
     )
diff --git a/egs/wenetspeech/ASR/pruned_transducer_stateless2/finetune.py b/egs/wenetspeech/ASR/pruned_transducer_stateless2/finetune.py
index ba91980d32..c34f1593d1 100755
--- a/egs/wenetspeech/ASR/pruned_transducer_stateless2/finetune.py
+++ b/egs/wenetspeech/ASR/pruned_transducer_stateless2/finetune.py
@@ -78,7 +78,7 @@ def add_finetune_arguments(parser: argparse.ArgumentParser):
         default=None,
         help="""
         Modules to be initialized. It matches all parameters starting with
-        a specific key. The keys are given with Comma seperated. If None,
+        a specific key. The keys are given with Comma separated. If None,
         all modules will be initialised. For example, if you only want to
         initialise all parameters staring with "encoder", use "encoder";
         if you want to initialise parameters starting with encoder or decoder,
diff --git a/icefall/utils.py b/icefall/utils.py
index 399e8d8b3c..a9e8a81b94 100644
--- a/icefall/utils.py
+++ b/icefall/utils.py
@@ -1977,7 +1977,7 @@ def parse_timestamps_and_texts(
         A k2.Fsa with best_paths.arcs.num_axes() == 3, i.e.
         containing multiple FSAs, which is expected to be the result
         of k2.shortest_path (otherwise the returned values won't
-        be meaningful). Attribtute `labels` is the prediction unit,
+        be meaningful). Attribute `labels` is the prediction unit,
         e.g., phone or BPE tokens. Attribute `aux_labels` is the word index.
       word_table:
         The word symbol table.