Refactor: split style_bert_vits2.nlp.chinese package

Configured so that the same public function is exported from the module with the same name for each language.

style_bert_vits2/nlp/__init__.py

@@ -82,7 +82,8 @@ def clean_text(
         norm_text = normalize_text(text)
         phones, tones, word2ph = g2p(norm_text)
     elif language == Languages.ZH:
-        from style_bert_vits2.nlp.chinese import g2p, normalize_text
+        from style_bert_vits2.nlp.chinese.g2p import g2p
+        from style_bert_vits2.nlp.chinese.normalizer import normalize_text
         norm_text = normalize_text(text)
         phones, tones, word2ph = g2p(norm_text)
     else:

style_bert_vits2/nlp/chinese/__init__.py

@@ -1,193 +0,0 @@
-import os
-import re
-
-import cn2an
-import jieba.posseg as psg
-from pypinyin import lazy_pinyin, Style
-
-from style_bert_vits2.nlp.chinese.tone_sandhi import ToneSandhi
-from style_bert_vits2.nlp.symbols import PUNCTUATIONS
-
-
-current_file_path = os.path.dirname(__file__)
-pinyin_to_symbol_map = {
-    line.split("\t")[0]: line.strip().split("\t")[1]
-    for line in open(os.path.join(current_file_path, "opencpop-strict.txt")).readlines()
-}
-
-
-rep_map = {
-    "：": ",",
-    "；": ",",
-    "，": ",",
-    "。": ".",
-    "！": "!",
-    "？": "?",
-    "\n": ".",
-    "·": ",",
-    "、": ",",
-    "...": "…",
-    "$": ".",
-    "“": "'",
-    "”": "'",
-    '"': "'",
-    "‘": "'",
-    "’": "'",
-    "（": "'",
-    "）": "'",
-    "(": "'",
-    ")": "'",
-    "《": "'",
-    "》": "'",
-    "【": "'",
-    "】": "'",
-    "[": "'",
-    "]": "'",
-    "—": "-",
-    "～": "-",
-    "~": "-",
-    "「": "'",
-    "」": "'",
-}
-
-tone_modifier = ToneSandhi()
-
-
-def replace_punctuation(text):
-    text = text.replace("嗯", "恩").replace("呣", "母")
-    pattern = re.compile("|".join(re.escape(p) for p in rep_map.keys()))
-
-    replaced_text = pattern.sub(lambda x: rep_map[x.group()], text)
-
-    replaced_text = re.sub(
-        r"[^\u4e00-\u9fa5" + "".join(PUNCTUATIONS) + r"]+", "", replaced_text
-    )
-
-    return replaced_text
-
-
-def g2p(text: str) -> tuple[list[str], list[int], list[int]]:
-    pattern = r"(?<=[{0}])\s*".format("".join(PUNCTUATIONS))
-    sentences = [i for i in re.split(pattern, text) if i.strip() != ""]
-    phones, tones, word2ph = _g2p(sentences)
-    assert sum(word2ph) == len(phones)
-    assert len(word2ph) == len(text)  # Sometimes it will crash,you can add a try-catch.
-    phones = ["_"] + phones + ["_"]
-    tones = [0] + tones + [0]
-    word2ph = [1] + word2ph + [1]
-    return phones, tones, word2ph
-
-
-def _get_initials_finals(word):
-    initials = []
-    finals = []
-    orig_initials = lazy_pinyin(word, neutral_tone_with_five=True, style=Style.INITIALS)
-    orig_finals = lazy_pinyin(
-        word, neutral_tone_with_five=True, style=Style.FINALS_TONE3
-    )
-    for c, v in zip(orig_initials, orig_finals):
-        initials.append(c)
-        finals.append(v)
-    return initials, finals
-
-
-def _g2p(segments):
-    phones_list = []
-    tones_list = []
-    word2ph = []
-    for seg in segments:
-        # Replace all English words in the sentence
-        seg = re.sub("[a-zA-Z]+", "", seg)
-        seg_cut = psg.lcut(seg)
-        initials = []
-        finals = []
-        seg_cut = tone_modifier.pre_merge_for_modify(seg_cut)
-        for word, pos in seg_cut:
-            if pos == "eng":
-                continue
-            sub_initials, sub_finals = _get_initials_finals(word)
-            sub_finals = tone_modifier.modified_tone(word, pos, sub_finals)
-            initials.append(sub_initials)
-            finals.append(sub_finals)
-
-            # assert len(sub_initials) == len(sub_finals) == len(word)
-        initials = sum(initials, [])
-        finals = sum(finals, [])
-        #
-        for c, v in zip(initials, finals):
-            raw_pinyin = c + v
-            # NOTE: post process for pypinyin outputs
-            # we discriminate i, ii and iii
-            if c == v:
-                assert c in PUNCTUATIONS
-                phone = [c]
-                tone = "0"
-                word2ph.append(1)
-            else:
-                v_without_tone = v[:-1]
-                tone = v[-1]
-
-                pinyin = c + v_without_tone
-                assert tone in "12345"
-
-                if c:
-                    # 多音节
-                    v_rep_map = {
-                        "uei": "ui",
-                        "iou": "iu",
-                        "uen": "un",
-                    }
-                    if v_without_tone in v_rep_map.keys():
-                        pinyin = c + v_rep_map[v_without_tone]
-                else:
-                    # 单音节
-                    pinyin_rep_map = {
-                        "ing": "ying",
-                        "i": "yi",
-                        "in": "yin",
-                        "u": "wu",
-                    }
-                    if pinyin in pinyin_rep_map.keys():
-                        pinyin = pinyin_rep_map[pinyin]
-                    else:
-                        single_rep_map = {
-                            "v": "yu",
-                            "e": "e",
-                            "i": "y",
-                            "u": "w",
-                        }
-                        if pinyin[0] in single_rep_map.keys():
-                            pinyin = single_rep_map[pinyin[0]] + pinyin[1:]
-
-                assert pinyin in pinyin_to_symbol_map.keys(), (pinyin, seg, raw_pinyin)
-                phone = pinyin_to_symbol_map[pinyin].split(" ")
-                word2ph.append(len(phone))
-
-            phones_list += phone
-            tones_list += [int(tone)] * len(phone)
-    return phones_list, tones_list, word2ph
-
-
-def normalize_text(text: str) -> str:
-    numbers = re.findall(r"\d+(?:\.?\d+)?", text)
-    for number in numbers:
-        text = text.replace(number, cn2an.an2cn(number), 1)
-    text = replace_punctuation(text)
-    return text
-
-
-if __name__ == "__main__":
-    from style_bert_vits2.nlp.chinese.bert_feature import extract_bert_feature
-
-    text = "啊！但是《原神》是由,米哈\游自主，  [研发]的一款全.新开放世界.冒险游戏"
-    text = normalize_text(text)
-    print(text)
-    phones, tones, word2ph = g2p(text)
-    bert = extract_bert_feature(text, word2ph, 'cuda')
-
-    print(phones, tones, word2ph, bert.shape)
-
-
-# # 示例用法
-# text = "这是一个示例文本：,你好！这是一个测试...."
-# print(g2p_paddle(text))  # 输出: 这是一个示例文本你好这是一个测试

style_bert_vits2/nlp/chinese/g2p.py

@@ -0,0 +1,135 @@
+import re
+from pathlib import Path
+
+import jieba.posseg as psg
+from pypinyin import lazy_pinyin, Style
+
+from style_bert_vits2.nlp.chinese.tone_sandhi import ToneSandhi
+from style_bert_vits2.nlp.symbols import PUNCTUATIONS
+
+
+__PINYIN_TO_SYMBOL_MAP = {
+    line.split("\t")[0]: line.strip().split("\t")[1]
+    for line in open(Path(__file__).parent / "opencpop-strict.txt").readlines()
+}
+
+
+def g2p(text: str) -> tuple[list[str], list[int], list[int]]:
+    pattern = r"(?<=[{0}])\s*".format("".join(PUNCTUATIONS))
+    sentences = [i for i in re.split(pattern, text) if i.strip() != ""]
+    phones, tones, word2ph = __g2p(sentences)
+    assert sum(word2ph) == len(phones)
+    assert len(word2ph) == len(text)  # Sometimes it will crash,you can add a try-catch.
+    phones = ["_"] + phones + ["_"]
+    tones = [0] + tones + [0]
+    word2ph = [1] + word2ph + [1]
+    return phones, tones, word2ph
+
+
+def __g2p(segments: list[str]) -> tuple[list[str], list[int], list[int]]:
+    phones_list = []
+    tones_list = []
+    word2ph = []
+    tone_modifier = ToneSandhi()
+    for seg in segments:
+        # Replace all English words in the sentence
+        seg = re.sub("[a-zA-Z]+", "", seg)
+        seg_cut = psg.lcut(seg)
+        initials = []
+        finals = []
+        seg_cut = tone_modifier.pre_merge_for_modify(seg_cut)  # type: ignore
+        for word, pos in seg_cut:
+            if pos == "eng":
+                continue
+            sub_initials, sub_finals = __get_initials_finals(word)
+            sub_finals = tone_modifier.modified_tone(word, pos, sub_finals)
+            initials.append(sub_initials)
+            finals.append(sub_finals)
+
+            # assert len(sub_initials) == len(sub_finals) == len(word)
+        initials = sum(initials, [])
+        finals = sum(finals, [])
+        #
+        for c, v in zip(initials, finals):
+            raw_pinyin = c + v
+            # NOTE: post process for pypinyin outputs
+            # we discriminate i, ii and iii
+            if c == v:
+                assert c in PUNCTUATIONS
+                phone = [c]
+                tone = "0"
+                word2ph.append(1)
+            else:
+                v_without_tone = v[:-1]
+                tone = v[-1]
+
+                pinyin = c + v_without_tone
+                assert tone in "12345"
+
+                if c:
+                    # 多音节
+                    v_rep_map = {
+                        "uei": "ui",
+                        "iou": "iu",
+                        "uen": "un",
+                    }
+                    if v_without_tone in v_rep_map.keys():
+                        pinyin = c + v_rep_map[v_without_tone]
+                else:
+                    # 单音节
+                    pinyin_rep_map = {
+                        "ing": "ying",
+                        "i": "yi",
+                        "in": "yin",
+                        "u": "wu",
+                    }
+                    if pinyin in pinyin_rep_map.keys():
+                        pinyin = pinyin_rep_map[pinyin]
+                    else:
+                        single_rep_map = {
+                            "v": "yu",
+                            "e": "e",
+                            "i": "y",
+                            "u": "w",
+                        }
+                        if pinyin[0] in single_rep_map.keys():
+                            pinyin = single_rep_map[pinyin[0]] + pinyin[1:]
+
+                assert pinyin in __PINYIN_TO_SYMBOL_MAP.keys(), (pinyin, seg, raw_pinyin)
+                phone = __PINYIN_TO_SYMBOL_MAP[pinyin].split(" ")
+                word2ph.append(len(phone))
+
+            phones_list += phone
+            tones_list += [int(tone)] * len(phone)
+    return phones_list, tones_list, word2ph
+
+
+def __get_initials_finals(word: str) -> tuple[list[str], list[str]]:
+    initials = []
+    finals = []
+    orig_initials = lazy_pinyin(word, neutral_tone_with_five=True, style=Style.INITIALS)
+    orig_finals = lazy_pinyin(
+        word, neutral_tone_with_five=True, style=Style.FINALS_TONE3
+    )
+    for c, v in zip(orig_initials, orig_finals):
+        initials.append(c)
+        finals.append(v)
+    return initials, finals
+
+
+if __name__ == "__main__":
+    from style_bert_vits2.nlp.chinese.bert_feature import extract_bert_feature
+    from style_bert_vits2.nlp.chinese.normalizer import normalize_text
+
+    text = "啊！但是《原神》是由,米哈游自主，  [研发]的一款全.新开放世界.冒险游戏"
+    text = normalize_text(text)
+    print(text)
+    phones, tones, word2ph = g2p(text)
+    bert = extract_bert_feature(text, word2ph, 'cuda')
+
+    print(phones, tones, word2ph, bert.shape)
+
+
+# 示例用法
+# text = "这是一个示例文本：,你好！这是一个测试...."
+# print(g2p_paddle(text))  # 输出: 这是一个示例文本你好这是一个测试

style_bert_vits2/nlp/chinese/normalizer.py

@@ -0,0 +1,61 @@
+import re
+
+import cn2an
+
+from style_bert_vits2.nlp.symbols import PUNCTUATIONS
+
+
+def normalize_text(text: str) -> str:
+    numbers = re.findall(r"\d+(?:\.?\d+)?", text)
+    for number in numbers:
+        text = text.replace(number, cn2an.an2cn(number), 1)
+    text = replace_punctuation(text)
+    return text
+
+
+def replace_punctuation(text: str) -> str:
+
+    REPLACE_MAP = {
+        "：": ",",
+        "；": ",",
+        "，": ",",
+        "。": ".",
+        "！": "!",
+        "？": "?",
+        "\n": ".",
+        "·": ",",
+        "、": ",",
+        "...": "…",
+        "$": ".",
+        "“": "'",
+        "”": "'",
+        '"': "'",
+        "‘": "'",
+        "’": "'",
+        "（": "'",
+        "）": "'",
+        "(": "'",
+        ")": "'",
+        "《": "'",
+        "》": "'",
+        "【": "'",
+        "】": "'",
+        "[": "'",
+        "]": "'",
+        "—": "-",
+        "～": "-",
+        "~": "-",
+        "「": "'",
+        "」": "'",
+    }
+
+    text = text.replace("嗯", "恩").replace("呣", "母")
+    pattern = re.compile("|".join(re.escape(p) for p in REPLACE_MAP.keys()))
+
+    replaced_text = pattern.sub(lambda x: REPLACE_MAP[x.group()], text)
+
+    replaced_text = re.sub(
+        r"[^\u4e00-\u9fa5" + "".join(PUNCTUATIONS) + r"]+", "", replaced_text
+    )
+
+    return replaced_text

style_bert_vits2/nlp/english/normalizer.py

@@ -14,22 +14,12 @@ __NUMBER_PATTERN = re.compile(r"[0-9]+")
 
 def normalize_text(text: str) -> str:
     text = __normalize_numbers(text)
-    text = __replace_punctuation(text)
+    text = replace_punctuation(text)
     text = re.sub(r"([,;.\?\!])([\w])", r"\1 \2", text)
     return text
 
 
-def __normalize_numbers(text: str) -> str:
+def replace_punctuation(text: str) -> str:
-    text = re.sub(__COMMA_NUMBER_PATTERN, __remove_commas, text)
-    text = re.sub(__POUNDS_PATTERN, r"\1 pounds", text)
-    text = re.sub(__DOLLARS_PATTERN, __expand_dollars, text)
-    text = re.sub(__DECIMAL_NUMBER_PATTERN, __expand_decimal_point, text)
-    text = re.sub(__ORDINAL_PATTERN, __expand_ordinal, text)
-    text = re.sub(__NUMBER_PATTERN, __expand_number, text)
-    return text
-
-
-def __replace_punctuation(text: str) -> str:
     REPLACE_MAP = {
         "：": ",",
         "；": ",",
@@ -80,6 +70,16 @@ def __replace_punctuation(text: str) -> str:
     return replaced_text
 
 
+def __normalize_numbers(text: str) -> str:
+    text = re.sub(__COMMA_NUMBER_PATTERN, __remove_commas, text)
+    text = re.sub(__POUNDS_PATTERN, r"\1 pounds", text)
+    text = re.sub(__DOLLARS_PATTERN, __expand_dollars, text)
+    text = re.sub(__DECIMAL_NUMBER_PATTERN, __expand_decimal_point, text)
+    text = re.sub(__ORDINAL_PATTERN, __expand_ordinal, text)
+    text = re.sub(__NUMBER_PATTERN, __expand_number, text)
+    return text
+
+
 def __expand_dollars(m: re.Match[str]) -> str:
     match = m.group(1)
     parts = match.split(".")