From dad6a7b0b6bba9cf5cc0c3c72c7b29e0905609db Mon Sep 17 00:00:00 2001
From: Julian Weber <julian.weber@hotmail.fr>
Date: Sat, 21 Oct 2023 11:26:03 +0200
Subject: [PATCH] Preserve [ja] token of the text processing

---
 TTS/tts/layers/xtts/tokenizer.py | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/TTS/tts/layers/xtts/tokenizer.py b/TTS/tts/layers/xtts/tokenizer.py
index 1d4ed235ea..4b9fb9edef 100644
--- a/TTS/tts/layers/xtts/tokenizer.py
+++ b/TTS/tts/layers/xtts/tokenizer.py
@@ -483,10 +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