fix bugs

GPT_SoVITS/AR/data/dataset.py

@@ -15,6 +15,8 @@ from torch.utils.data import DataLoader
 from torch.utils.data import Dataset
 from transformers import AutoTokenizer
 
+version = os.environ.get('version',None)
+
 from text import cleaned_text_to_sequence
 
 # from config import exp_dir
@@ -149,7 +151,7 @@ class Text2SemanticDataset(Dataset):
             phoneme = phoneme.split(" ")
 
             try:
-                phoneme_ids = cleaned_text_to_sequence(phoneme)
+                phoneme_ids = cleaned_text_to_sequence(phoneme, version)
             except:
                 traceback.print_exc()
                 # print(f"{item_name} not in self.phoneme_data !")

GPT_SoVITS/module/data_utils.py

@@ -18,7 +18,7 @@ import requests
 from scipy.io import wavfile
 from io import BytesIO
 from tools.my_utils import load_audio
-
+version = os.environ.get('version',None)
 # ZeroDivisionError fixed by Tybost (https://github.com/RVC-Boss/GPT-SoVITS/issues/79)
 class TextAudioSpeakerLoader(torch.utils.data.Dataset):
     """
@@ -77,7 +77,7 @@ class TextAudioSpeakerLoader(torch.utils.data.Dataset):
             try:
                 phoneme = self.phoneme_data[audiopath][0]
                 phoneme = phoneme.split(' ')
-                phoneme_ids = cleaned_text_to_sequence(phoneme)
+                phoneme_ids = cleaned_text_to_sequence(phoneme, version)
             except Exception:
                 print(f"{audiopath} not in self.phoneme_data !")
                 skipped_phone += 1

GPT_SoVITS/text/cleaner.py

@@ -84,9 +84,11 @@ def clean_special(text, language, special_s, target_symbol, version=None):
     return new_ph, phones[1], norm_text
 
 
-def text_to_sequence(text, language):
+def text_to_sequence(text, language, version=None):
+    version = os.environ.get('version',version)
+    if version is None:version='v2'
     phones = clean_text(text)
-    return cleaned_text_to_sequence(phones)
+    return cleaned_text_to_sequence(phones, version)
 
 
 if __name__ == "__main__":