Refactor: minor adjustments

style_bert_vits2/text_processing/chinese/bert_feature.py

@@ -7,7 +7,7 @@ from style_bert_vits2.constants import Languages
 from style_bert_vits2.text_processing import bert_models
 
 
-models: dict[str, PreTrainedModel] = {}
+models: dict[torch.device | str, PreTrainedModel] = {}
 
 
 def extract_bert_feature(

style_bert_vits2/text_processing/english/bert_feature.py

@@ -7,7 +7,7 @@ from style_bert_vits2.constants import Languages
 from style_bert_vits2.text_processing import bert_models
 
 
-models: dict[str, PreTrainedModel] = {}
+models: dict[torch.device | str, PreTrainedModel] = {}
 
 
 def extract_bert_feature(

style_bert_vits2/text_processing/japanese/bert_feature.py

@@ -8,7 +8,7 @@ from style_bert_vits2.text_processing import bert_models
 from style_bert_vits2.text_processing.japanese.g2p import text_to_sep_kata
 
 
-models: dict[str, PreTrainedModel] = {}
+models: dict[torch.device | str, PreTrainedModel] = {}
 
 
 def extract_bert_feature(

style_bert_vits2/text_processing/japanese/g2p_utils.py

@@ -9,7 +9,7 @@ from style_bert_vits2.text_processing.symbols import PUNCTUATIONS
 def g2kata_tone(norm_text: str) -> list[tuple[str, int]]:
     """
     テキストからカタカナとアクセントのペアのリストを返す。
-    推論時のみに使われるので、常に `raise_yomi_error=False` で g2p を呼ぶ。
+    推論時のみに使われる関数のため、常に `raise_yomi_error=False` を指定して g2p() を呼ぶ仕様になっている。
 
     Args:
         norm_text: 正規化されたテキスト。

style_bert_vits2/text_processing/symbols.py

@@ -77,8 +77,8 @@ ZH_SYMBOLS = [
 ]
 NUM_ZH_TONES = 6
 
-# japanese
+# Japanese
-JA_SYMBOLS = [
+JP_SYMBOLS = [
     "N",
     "a",
     "a:",
@@ -122,7 +122,7 @@ JA_SYMBOLS = [
     "z",
     "zy",
 ]
-NUM_JA_TONES = 2
+NUM_JP_TONES = 2
 
 # English
 EN_SYMBOLS = [
@@ -169,23 +169,25 @@ EN_SYMBOLS = [
 NUM_EN_TONES = 4
 
 # Combine all symbols
-NORMAL_SYMBOLS = sorted(set(ZH_SYMBOLS + JA_SYMBOLS + EN_SYMBOLS))
+NORMAL_SYMBOLS = sorted(set(ZH_SYMBOLS + JP_SYMBOLS + EN_SYMBOLS))
 SYMBOLS = [PAD] + NORMAL_SYMBOLS + PUNCTUATION_SYMBOLS
 SIL_PHONEMES_IDS = [SYMBOLS.index(i) for i in PUNCTUATION_SYMBOLS]
 
 # Combine all tones
-NUM_TONES = NUM_ZH_TONES + NUM_JA_TONES + NUM_EN_TONES
+NUM_TONES = NUM_ZH_TONES + NUM_JP_TONES + NUM_EN_TONES
 
 # Language maps
 LANGUAGE_ID_MAP = {"ZH": 0, "JP": 1, "EN": 2}
 NUM_LANGUAGES = len(LANGUAGE_ID_MAP.keys())
 
+# Language tone start map
 LANGUAGE_TONE_START_MAP = {
     "ZH": 0,
     "JP": NUM_ZH_TONES,
-    "EN": NUM_ZH_TONES + NUM_JA_TONES,
+    "EN": NUM_ZH_TONES + NUM_JP_TONES,
 }
 
+
 if __name__ == "__main__":
     a = set(ZH_SYMBOLS)
     b = set(EN_SYMBOLS)

style_bert_vits2/utils/subprocess.py

@@ -5,8 +5,6 @@ from typing import Any, Callable
 from style_bert_vits2.logging import logger
 from style_bert_vits2.utils.stdout_wrapper import SAFE_STDOUT
 
-PYTHON = sys.executable
-
 
 def run_script_with_log(cmd: list[str], ignore_warning: bool = False) -> tuple[bool, str]:
     """
@@ -22,7 +20,7 @@ def run_script_with_log(cmd: list[str], ignore_warning: bool = False) -> tuple[b
 
     logger.info(f"Running: {' '.join(cmd)}")
     result = subprocess.run(
-        [PYTHON] + cmd,
+        [sys.executable] + cmd,
         stdout = SAFE_STDOUT,
         stderr = subprocess.PIPE,
         text = True,

utils.py

@@ -450,21 +450,6 @@ class HParams:
         return self.__dict__.__repr__()
 
 
-def load_model(model_path, config_path):
-    hps = get_hparams_from_file(config_path)
-    net = SynthesizerTrn(
-        # len(symbols),
-        108,
-        hps.data.filter_length // 2 + 1,
-        hps.train.segment_size // hps.data.hop_length,
-        n_speakers=hps.data.n_speakers,
-        **hps.model,
-    ).to("cpu")
-    _ = net.eval()
-    _ = load_checkpoint(model_path, net, None, skip_optimizer=True)
-    return net
-
-
 def mix_model(
     network1, network2, output_path, voice_ratio=(0.5, 0.5), tone_ratio=(0.5, 0.5)
 ):