diff --git a/TTS/tts/layers/xtts/tokenizer.py b/TTS/tts/layers/xtts/tokenizer.py
index f34a7ac036..4b9fb9edef 100644
--- a/TTS/tts/layers/xtts/tokenizer.py
+++ b/TTS/tts/layers/xtts/tokenizer.py
@@ -6,7 +6,6 @@
 from tokenizers import Tokenizer
 
 import pypinyin
-import cutlet
 from num2words import num2words
 from TTS.tts.layers.xtts.zh_num2words import TextNorm as zh_num2words
 
@@ -484,10 +483,13 @@ def preprocess_text(self, txt, lang):
             if lang == "zh-cn":
                 txt = chinese_transliterate(txt)
         elif lang == "ja":
+            assert txt[:4] == "[ja]", "Japanese speech should start with the [ja] token."
+            txt = txt[4:]
             if self.katsu is None:
                 import cutlet
                 self.katsu = cutlet.Cutlet()
             txt = japanese_cleaners(txt, self.katsu)
+            txt = "[ja]" + txt
         else:
             raise NotImplementedError()
         return txt