diff --git a/wenet/dataset/processor.py b/wenet/dataset/processor.py
index 68e500e45..a769eba8e 100644
--- a/wenet/dataset/processor.py
+++ b/wenet/dataset/processor.py
@@ -27,7 +27,7 @@
 from torch.nn.utils.rnn import pad_sequence
 from wenet.text.base_tokenizer import BaseTokenizer
 
-# torchaudio.utils.sox_utils.set_buffer_size(16500)
+torchaudio.utils.sox_utils.set_buffer_size(16500)
 
 AUDIO_FORMAT_SETS = set(['flac', 'mp3', 'm4a', 'ogg', 'opus', 'wav', 'wma'])
 
diff --git a/wenet/utils/common.py b/wenet/utils/common.py
index 7384bad57..5da1fb341 100644
--- a/wenet/utils/common.py
+++ b/wenet/utils/common.py
@@ -20,6 +20,9 @@
 import torch
 from torch.nn.utils.rnn import pad_sequence
 
+from whisper.tokenizer import LANGUAGES as WhiserLanguages
+
+WHISPER_LANGS = tuple(WhiserLanguages.keys())
 IGNORE_ID = -1
 
 
@@ -173,8 +176,6 @@ def add_whisper_tokens(
         ys_out (torch.Tensor) : (B, Lmax + ?)
 
     """
-    from whisper.tokenizer import LANGUAGES as WhiserLanguages
-    WHISPER_LANGS = tuple(WhiserLanguages.keys())
     if use_prev:
         # i.e., hotword list
         _prev = [special_tokens["sot_prev"]]