Update Bert-VITS2 num_tones

bert_vits2/bert_vits2.py

@@ -27,12 +27,14 @@ class Bert_VITS2:
         self.bert_model_names = {"zh": "CHINESE_ROBERTA_WWM_EXT_LARGE"}
         self.ja_bert_dim = 1024
         self.ja_extra_str = ""
+        self.num_tones = num_tones
 
         if self.version in ["1.0", "1.0.0", "1.0.1"]:
             self.symbols = symbols_legacy
             self.hps_ms.model.n_layers_trans_flow = 3
             self.lang = ["zh"]
             self.ja_bert_dim = 768
+            self.num_tones = num_tones_v111
 
         elif self.version in ["1.1.0-transition"]:
             self.hps_ms.model.n_layers_trans_flow = 3
@@ -40,6 +42,7 @@ class Bert_VITS2:
             self.bert_model_names["ja"] = "BERT_BASE_JAPANESE_V3"
             self.ja_bert_dim = 768
             self.ja_extra_str = "_v111"
+            self.num_tones = num_tones_v111
 
         elif self.version in ["1.1", "1.1.0", "1.1.1"]:
             self.hps_ms.model.n_layers_trans_flow = 6
@@ -47,12 +50,15 @@ class Bert_VITS2:
             self.bert_model_names["ja"] = "BERT_BASE_JAPANESE_V3"
             self.ja_bert_dim = 768
             self.ja_extra_str = "_v111"
+            self.num_tones = num_tones_v111
 
         elif self.version in ["2.0", "2.0.0"]:
             self.hps_ms.model.n_layers_trans_flow = 4
             self.bert_model_names = {"zh": "CHINESE_ROBERTA_WWM_EXT_LARGE",
                                      "ja": "DEBERTA_V2_LARGE_JAPANESE",
                                      "en": "DEBERTA_V3_LARGE"}
+            self.num_tones = num_tones
+            
 
         # self.bert_handler = BertHandler(self.lang)
 
@@ -69,6 +75,7 @@ class Bert_VITS2:
             n_speakers=self.hps_ms.data.n_speakers,
             symbols=self.symbols,
             ja_bert_dim=self.ja_bert_dim,
+            num_tones=self.num_tones,
             **self.hps_ms.model).to(self.device)
         _ = self.net_g.eval()
         bert_vits2_utils.load_checkpoint(self.model_path, self.net_g, None, skip_optimizer=True, version=self.version)

bert_vits2/models.py

@@ -11,7 +11,7 @@ from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
 from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
 
 from bert_vits2.commons import init_weights, get_padding
-from bert_vits2.text import num_tones, num_languages
+from bert_vits2.text import num_languages
 
 
 class DurationDiscriminator(nn.Module):  # vits2
@@ -258,7 +258,8 @@ class TextEncoder(nn.Module):
                  p_dropout,
                  gin_channels=0,
                  symbols=None,
-                 ja_bert_dim=1024):
+                 ja_bert_dim=1024,
+                 num_tones=None):
         super().__init__()
         self.n_vocab = n_vocab
         self.out_channels = out_channels
@@ -601,6 +602,7 @@ class SynthesizerTrn(nn.Module):
                  use_transformer_flow=True,
                  symbols=None,
                  ja_bert_dim=1024,
+                 num_tones=None,
                  **kwargs):
 
         super().__init__()
@@ -641,7 +643,8 @@ class SynthesizerTrn(nn.Module):
                                  p_dropout,
                                  gin_channels=self.enc_gin_channels,
                                  symbols=symbols,
-                                 ja_bert_dim=ja_bert_dim
+                                 ja_bert_dim=ja_bert_dim,
+                                 num_tones=num_tones
                                  )
         self.dec = Generator(inter_channels, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates,
                              upsample_initial_channel, upsample_kernel_sizes, gin_channels=gin_channels)

bert_vits2/text/__init__.py

@@ -2,6 +2,15 @@ from bert_vits2.text.symbols import *
 from bert_vits2.text.bert_handler import BertHandler
 
 
+def cleaned_text_to_sequence_v111(cleaned_text, tones, language, _symbol_to_id):
+    """version <= 1.1.1"""
+    phones = [_symbol_to_id[symbol] for symbol in cleaned_text]
+    tone_start = language_tone_start_map_v111[language]
+    tones = [i + tone_start for i in tones]
+    lang_id = language_id_map[language]
+    lang_ids = [lang_id for i in phones]
+    return phones, tones, lang_ids
+
 def cleaned_text_to_sequence(cleaned_text, tones, language, _symbol_to_id):
     """Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
     Args:

bert_vits2/text/symbols.py

@@ -120,7 +120,8 @@ ja_symbols = [
     "z",
     "zy",
 ]
-num_ja_tones = 1
+num_ja_tones_v111 = 1
+num_ja_tones = 2
 
 # English
 en_symbols = [
@@ -176,12 +177,19 @@ symbols_legacy = [pad] + normal_symbols_legacy + pu_symbols
 sil_phonemes_ids_legacy = [symbols_legacy.index(i) for i in pu_symbols]
 
 # combine all tones
+num_tones_v111 = num_zh_tones + num_ja_tones_v111 + num_en_tones
 num_tones = num_zh_tones + num_ja_tones + num_en_tones
 
 # language maps
 language_id_map = {"zh": 0, "ja": 1, "en": 2}
 num_languages = len(language_id_map.keys())
 
+language_tone_start_map_v111 = {
+    "zh": 0,
+    "ja": num_zh_tones,
+    "en": num_zh_tones + num_ja_tones_v111,
+}
+
 language_tone_start_map = {
     "zh": 0,
     "ja": num_zh_tones,