Upload 25 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+text/lexicon/zaonhe.ocd2 filter=lfs diff=lfs merge=lfs -text

text/__init__.py

@@ -0,0 +1,28 @@
+from text.symbols import *
+
+_symbol_to_id = {s: i for i, s in enumerate(symbols)}
+
+
+def cleaned_text_to_sequence(cleaned_text, tones, language):
+    """Converts a string of text to a sequence of IDs corresponding to the symbols in the text.
+    Args:
+      text: string to convert to a sequence
+    Returns:
+      List of integers corresponding to the symbols in the text
+    """
+    phones = [_symbol_to_id[symbol] for symbol in cleaned_text]
+    tone_start = language_tone_start_map[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 get_bert(norm_text, word2ph, language, device):
+    from .chinese_bert import get_bert_feature as zh_bert
+    from .english_bert_mock import get_bert_feature as en_bert
+    from .shanghainese_bert import get_bert_feature as sh_bert
+
+    lang_bert_func_map = {"ZH": zh_bert, "EN": en_bert, "SH": sh_bert}
+    bert = lang_bert_func_map[language](norm_text, word2ph, device)
+    return bert

text/chinese.py

@@ -0,0 +1,193 @@
+import os
+import re
+
+import cn2an
+from pypinyin import lazy_pinyin, Style
+
+from text.symbols import punctuation, zh_symbols
+from text.tone_sandhi import ToneSandhi
+
+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()
+}
+
+import jieba.posseg as psg
+
+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(punctuation) + r"]+", "", replaced_text
+    )
+
+    return replaced_text
+
+
+def g2p(text):
+    pattern = r"(?<=[{0}])\s*".format("".join(punctuation))
+    sentences = [i for i in re.split(pattern, text) if i.strip() != ""]
+    phones, tones, word2ph = _g2p(sentences)
+    # phones 加前缀 'ZH'
+    phones = ["ZH_" + i if ("ZH_" + i) in zh_symbols else i for i in phones]
+    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 punctuation
+                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 text_normalize(text):
+    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 get_bert_feature(text, word2ph, device):
+    from text import chinese_bert
+    return chinese_bert.get_bert_feature(text, word2ph, device)
+
+
+if __name__ == "__main__":
+    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 = text_normalize(text)
+    print(text)
+    phones, tones, word2ph = g2p(text)
+    print(phones, tones, word2ph)
+
+# # 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

text/chinese_bert.py

@@ -0,0 +1,99 @@
+import sys
+
+import torch
+from transformers import AutoTokenizer, AutoModelForMaskedLM
+
+model_name = "./bert/chinese-roberta-wwm-ext-large"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+models = dict()
+
+
+def get_bert_feature(text, word2ph, device="cuda"):
+    if (
+            sys.platform == "darwin"
+            and torch.backends.mps.is_available()
+            and device == "cpu"
+    ):
+        device = "mps"
+    if not device:
+        device = "cuda"
+    if device not in models.keys():
+        models[device] = AutoModelForMaskedLM.from_pretrained(model_name).to(device)
+
+    with torch.no_grad():
+        inputs = tokenizer(text, return_tensors="pt")
+        for i in inputs:
+            inputs[i] = inputs[i].to(device)
+        res = models[device](**inputs, output_hidden_states=True)
+        res = torch.cat(res["hidden_states"][-3:-2], -1)[0].cpu()
+
+    assert len(word2ph) == len(text) + 2
+    word2phone = word2ph
+    phone_level_feature = []
+    for i in range(len(word2phone)):
+        repeat_feature = res[i].repeat(word2phone[i], 1)
+        phone_level_feature.append(repeat_feature)
+
+    phone_level_feature = torch.cat(phone_level_feature, dim=0)
+
+    return phone_level_feature.T
+
+
+if __name__ == "__main__":
+    word_level_feature = torch.rand(38, 1024)  # 12个词,每个词1024维特征
+    word2phone = [
+        1,
+        2,
+        1,
+        2,
+        2,
+        1,
+        2,
+        2,
+        1,
+        2,
+        2,
+        1,
+        2,
+        2,
+        2,
+        2,
+        2,
+        1,
+        1,
+        2,
+        2,
+        1,
+        2,
+        2,
+        2,
+        2,
+        1,
+        2,
+        2,
+        2,
+        2,
+        2,
+        1,
+        2,
+        2,
+        2,
+        2,
+        1,
+    ]
+
+    # 计算总帧数
+    total_frames = sum(word2phone)
+    print(word_level_feature.shape)
+    print(word2phone)
+    phone_level_feature = []
+    for i in range(len(word2phone)):
+        print(word_level_feature[i].shape)
+
+        # 对每个词重复word2phone[i]次
+        repeat_feature = word_level_feature[i].repeat(word2phone[i], 1)
+        phone_level_feature.append(repeat_feature)
+
+    phone_level_feature = torch.cat(phone_level_feature, dim=0)
+    print(phone_level_feature.shape)  # torch.Size([36, 1024])

text/cleaner.py

@@ -0,0 +1,28 @@
+from text import chinese, shanghainese, english, cleaned_text_to_sequence
+
+
+language_module_map = {"ZH": chinese, "SH": shanghainese, "EN": english}
+
+
+def clean_text(text, language):
+    language_module = language_module_map[language]
+    norm_text = language_module.text_normalize(text)
+    phones, tones, word2ph = language_module.g2p(norm_text)
+    return norm_text, phones, tones, word2ph
+
+
+def clean_text_bert(text, language, device):
+    language_module = language_module_map[language]
+    norm_text = language_module.text_normalize(text)
+    phones, tones, word2ph = language_module.g2p(norm_text)
+    bert = language_module.get_bert_feature(norm_text, word2ph, device)
+    return phones, tones, bert
+
+
+def text_to_sequence(text, language):
+    norm_text, phones, tones, word2ph = clean_text(text, language)
+    return cleaned_text_to_sequence(phones, tones, language)
+
+
+if __name__ == "__main__":
+    pass

text/cmudict_cache.pickle

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9b21b20325471934ba92f2e4a5976989e7d920caa32e7a286eacb027d197949
+size 6212655

text/english.py

@@ -0,0 +1,348 @@
+import os
+import pickle
+
+from g2p_en import G2p
+
+from text import symbols
+
+current_file_path = os.path.dirname(__file__)
+CMU_DICT_PATH = os.path.join(current_file_path, "cmudict.rep")
+CACHE_PATH = os.path.join(current_file_path, "cmudict_cache.pickle")
+_g2p = G2p()
+
+arpa = {
+    "AH0",
+    "S",
+    "AH1",
+    "EY2",
+    "AE2",
+    "EH0",
+    "OW2",
+    "UH0",
+    "NG",
+    "B",
+    "G",
+    "AY0",
+    "M",
+    "AA0",
+    "F",
+    "AO0",
+    "ER2",
+    "UH1",
+    "IY1",
+    "AH2",
+    "DH",
+    "IY0",
+    "EY1",
+    "IH0",
+    "K",
+    "N",
+    "W",
+    "IY2",
+    "T",
+    "AA1",
+    "ER1",
+    "EH2",
+    "OY0",
+    "UH2",
+    "UW1",
+    "Z",
+    "AW2",
+    "AW1",
+    "V",
+    "UW2",
+    "AA2",
+    "ER",
+    "AW0",
+    "UW0",
+    "R",
+    "OW1",
+    "EH1",
+    "ZH",
+    "AE0",
+    "IH2",
+    "IH",
+    "Y",
+    "JH",
+    "P",
+    "AY1",
+    "EY0",
+    "OY2",
+    "TH",
+    "HH",
+    "D",
+    "ER0",
+    "CH",
+    "AO1",
+    "AE1",
+    "AO2",
+    "OY1",
+    "AY2",
+    "IH1",
+    "OW0",
+    "L",
+    "SH",
+}
+
+
+def post_replace_ph(ph):
+    rep_map = {
+        "：": ",",
+        "；": ",",
+        "，": ",",
+        "。": ".",
+        "！": "!",
+        "？": "?",
+        "\n": ".",
+        "·": ",",
+        "、": ",",
+        "...": "…",
+        "v": "V",
+    }
+    if ph in rep_map.keys():
+        ph = rep_map[ph]
+    if ph in symbols:
+        return ph
+    ph = 'EN_' + ph
+    if ph not in symbols:
+        ph = "UNK"
+    return ph
+
+
+def read_dict():
+    g2p_dict = {}
+    start_line = 49
+    with open(CMU_DICT_PATH) as f:
+        line = f.readline()
+        line_index = 1
+        while line:
+            if line_index >= start_line:
+                line = line.strip()
+                word_split = line.split("  ")
+                word = word_split[0]
+
+                syllable_split = word_split[1].split(" - ")
+                g2p_dict[word] = []
+                for syllable in syllable_split:
+                    phone_split = syllable.split(" ")
+                    g2p_dict[word].append(phone_split)
+
+            line_index = line_index + 1
+            line = f.readline()
+
+    return g2p_dict
+
+
+def cache_dict(g2p_dict, file_path):
+    with open(file_path, "wb") as pickle_file:
+        pickle.dump(g2p_dict, pickle_file)
+
+
+def get_dict():
+    if os.path.exists(CACHE_PATH):
+        with open(CACHE_PATH, "rb") as pickle_file:
+            g2p_dict = pickle.load(pickle_file)
+    else:
+        g2p_dict = read_dict()
+        cache_dict(g2p_dict, CACHE_PATH)
+
+    return g2p_dict
+
+
+eng_dict = get_dict()
+
+
+def refine_ph(phn):
+    tone = 0
+    if re.search(r"\d$", phn):
+        tone = int(phn[-1]) + 1
+        phn = phn[:-1]
+    return phn.lower(), tone
+
+
+def refine_syllables(syllables):
+    tones = []
+    phonemes = []
+    for phn_list in syllables:
+        for i in range(len(phn_list)):
+            phn = phn_list[i]
+            phn, tone = refine_ph(phn)
+            phonemes.append(phn)
+            tones.append(tone)
+    return phonemes, tones
+
+
+import re
+import inflect
+
+_inflect = inflect.engine()
+_comma_number_re = re.compile(r"([0-9][0-9\,]+[0-9])")
+_decimal_number_re = re.compile(r"([0-9]+\.[0-9]+)")
+_pounds_re = re.compile(r"£([0-9\,]*[0-9]+)")
+_dollars_re = re.compile(r"\$([0-9\.\,]*[0-9]+)")
+_ordinal_re = re.compile(r"[0-9]+(st|nd|rd|th)")
+_number_re = re.compile(r"[0-9]+")
+
+# List of (regular expression, replacement) pairs for abbreviations:
+_abbreviations = [
+    (re.compile("\\b%s\\." % x[0], re.IGNORECASE), x[1])
+    for x in [
+        ("mrs", "misess"),
+        ("mr", "mister"),
+        ("dr", "doctor"),
+        ("st", "saint"),
+        ("co", "company"),
+        ("jr", "junior"),
+        ("maj", "major"),
+        ("gen", "general"),
+        ("drs", "doctors"),
+        ("rev", "reverend"),
+        ("lt", "lieutenant"),
+        ("hon", "honorable"),
+        ("sgt", "sergeant"),
+        ("capt", "captain"),
+        ("esq", "esquire"),
+        ("ltd", "limited"),
+        ("col", "colonel"),
+        ("ft", "fort"),
+    ]
+]
+
+# List of (ipa, lazy ipa) pairs:
+_lazy_ipa = [
+    (re.compile("%s" % x[0]), x[1])
+    for x in [
+        ("r", "ɹ"),
+        ("æ", "e"),
+        ("ɑ", "a"),
+        ("ɔ", "o"),
+        ("ð", "z"),
+        ("θ", "s"),
+        ("ɛ", "e"),
+        ("ɪ", "i"),
+        ("ʊ", "u"),
+        ("ʒ", "ʥ"),
+        ("ʤ", "ʥ"),
+        ("ˈ", "↓"),
+    ]
+]
+
+# List of (ipa, lazy ipa2) pairs:
+_lazy_ipa2 = [
+    (re.compile("%s" % x[0]), x[1])
+    for x in [
+        ("r", "ɹ"),
+        ("ð", "z"),
+        ("θ", "s"),
+        ("ʒ", "ʑ"),
+        ("ʤ", "dʑ"),
+        ("ˈ", "↓"),
+    ]
+]
+
+# List of (ipa, ipa2) pairs
+_ipa_to_ipa2 = [
+    (re.compile("%s" % x[0]), x[1]) for x in [("r", "ɹ"), ("ʤ", "dʒ"), ("ʧ", "tʃ")]
+]
+
+
+def _expand_dollars(m):
+    match = m.group(1)
+    parts = match.split(".")
+    if len(parts) > 2:
+        return match + " dollars"  # Unexpected format
+    dollars = int(parts[0]) if parts[0] else 0
+    cents = int(parts[1]) if len(parts) > 1 and parts[1] else 0
+    if dollars and cents:
+        dollar_unit = "dollar" if dollars == 1 else "dollars"
+        cent_unit = "cent" if cents == 1 else "cents"
+        return "%s %s, %s %s" % (dollars, dollar_unit, cents, cent_unit)
+    elif dollars:
+        dollar_unit = "dollar" if dollars == 1 else "dollars"
+        return "%s %s" % (dollars, dollar_unit)
+    elif cents:
+        cent_unit = "cent" if cents == 1 else "cents"
+        return "%s %s" % (cents, cent_unit)
+    else:
+        return "zero dollars"
+
+
+def _remove_commas(m):
+    return m.group(1).replace(",", "")
+
+
+def _expand_ordinal(m):
+    return _inflect.number_to_words(m.group(0))
+
+
+def _expand_number(m):
+    num = int(m.group(0))
+    if num > 1000 and num < 3000:
+        if num == 2000:
+            return "two thousand"
+        elif num > 2000 and num < 2010:
+            return "two thousand " + _inflect.number_to_words(num % 100)
+        elif num % 100 == 0:
+            return _inflect.number_to_words(num // 100) + " hundred"
+        else:
+            return _inflect.number_to_words(
+                num, andword="", zero="oh", group=2
+            ).replace(", ", " ")
+    else:
+        return _inflect.number_to_words(num, andword="")
+
+
+def _expand_decimal_point(m):
+    return m.group(1).replace(".", " point ")
+
+
+def normalize_numbers(text):
+    text = re.sub(_comma_number_re, _remove_commas, text)
+    text = re.sub(_pounds_re, r"\1 pounds", text)
+    text = re.sub(_dollars_re, _expand_dollars, text)
+    text = re.sub(_decimal_number_re, _expand_decimal_point, text)
+    text = re.sub(_ordinal_re, _expand_ordinal, text)
+    text = re.sub(_number_re, _expand_number, text)
+    return text
+
+
+def text_normalize(text):
+    text = normalize_numbers(text)
+    return text
+
+
+def g2p(text):
+    phones = []
+    tones = []
+    word2ph = []
+    words = re.split(r"([,;.\-\?\!\s+])", text)
+    words = [word for word in words if word.strip() != ""]
+    for word in words:
+        if word.upper() in eng_dict:
+            phns, tns = refine_syllables(eng_dict[word.upper()])
+            phones += phns
+            tones += tns
+            word2ph.append(len(phns))
+        else:
+            phone_list = list(filter(lambda p: p != " ", _g2p(word)))
+            for ph in phone_list:
+                if ph in arpa:
+                    ph, tn = refine_ph(ph)
+                    phones.append(ph)
+                    tones.append(tn)
+                else:
+                    phones.append(ph)
+                    tones.append(0)
+            word2ph.append(len(phone_list))
+
+    phones = [post_replace_ph(i) for i in phones]
+    phones = ["_"] + phones + ["_"]
+    tones = [0] + tones + [0]
+    word2ph = [1] + word2ph + [1]
+
+    return phones, tones, word2ph
+
+
+def get_bert_feature(text, word2ph, device):
+    from text import english_bert_mock
+    return english_bert_mock.get_bert_feature(text, word2ph, device)

text/english_bert_mock.py

@@ -0,0 +1,39 @@
+import sys
+
+import torch
+from transformers import DebertaV2Model, DebertaV2Tokenizer
+
+tokenizer = DebertaV2Tokenizer.from_pretrained("./bert/deberta-v3-large")
+
+models = dict()
+
+
+def get_bert_feature(text, word2ph, device="cuda"):
+    if (
+            sys.platform == "darwin"
+            and torch.backends.mps.is_available()
+            and device == "cpu"
+    ):
+        device = "mps"
+    if not device:
+        device = "cuda"
+    if device not in models.keys():
+        models[device] = DebertaV2Model.from_pretrained("./bert/deberta-v3-large").to(
+            device
+        )
+    with torch.no_grad():
+        inputs = tokenizer(text, return_tensors="pt")
+        for i in inputs:
+            inputs[i] = inputs[i].to(device)
+        res = models[device](**inputs, output_hidden_states=True)
+        res = torch.cat(res["hidden_states"][-3:-2], -1)[0].cpu()
+    # assert len(word2ph) == len(text)+2
+    word2phone = word2ph
+    phone_level_feature = []
+    for i in range(len(word2phone)):
+        repeat_feature = res[i].repeat(word2phone[i], 1)
+        phone_level_feature.append(repeat_feature)
+
+    phone_level_feature = torch.cat(phone_level_feature, dim=0)
+
+    return phone_level_feature.T

text/lexicon/zaonhe.json

@@ -0,0 +1,23 @@
+{
+	"name": "Shanghainese to IPA",
+	"segmentation": {
+		"type": "mmseg",
+		"dict": {
+			"type": "ocd2",
+			"file": "zaonhe.ocd2"
+		}
+	},
+	"conversion_chain": [
+		{
+			"dict": {
+				"type": "group",
+				"dicts": [
+					{
+						"type": "ocd2",
+						"file": "zaonhe.ocd2"
+					}
+				]
+			}
+		}
+	]
+}

text/lexicon/zaonhe.ocd2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a71b5a97eb49699f440137391565d208ea82156f0765986b7f3e16909e15672e
+size 4095228

text/opencpop-strict.txt

@@ -0,0 +1,429 @@
+a	AA a
+ai	AA ai
+an	AA an
+ang	AA ang
+ao	AA ao
+ba	b a
+bai	b ai
+ban	b an
+bang	b ang
+bao	b ao
+bei	b ei
+ben	b en
+beng	b eng
+bi	b i
+bian	b ian
+biao	b iao
+bie	b ie
+bin	b in
+bing	b ing
+bo	b o
+bu	b u
+ca	c a
+cai	c ai
+can	c an
+cang	c ang
+cao	c ao
+ce	c e
+cei	c ei
+cen	c en
+ceng	c eng
+cha	ch a
+chai	ch ai
+chan	ch an
+chang	ch ang
+chao	ch ao
+che	ch e
+chen	ch en
+cheng	ch eng
+chi	ch ir
+chong	ch ong
+chou	ch ou
+chu	ch u
+chua	ch ua
+chuai	ch uai
+chuan	ch uan
+chuang	ch uang
+chui	ch ui
+chun	ch un
+chuo	ch uo
+ci	c i0
+cong	c ong
+cou	c ou
+cu	c u
+cuan	c uan
+cui	c ui
+cun	c un
+cuo	c uo
+da	d a
+dai	d ai
+dan	d an
+dang	d ang
+dao	d ao
+de	d e
+dei	d ei
+den	d en
+deng	d eng
+di	d i
+dia	d ia
+dian	d ian
+diao	d iao
+die	d ie
+ding	d ing
+diu	d iu
+dong	d ong
+dou	d ou
+du	d u
+duan	d uan
+dui	d ui
+dun	d un
+duo	d uo
+e	EE e
+ei	EE ei
+en	EE en
+eng	EE eng
+er	EE er
+fa	f a
+fan	f an
+fang	f ang
+fei	f ei
+fen	f en
+feng	f eng
+fo	f o
+fou	f ou
+fu	f u
+ga	g a
+gai	g ai
+gan	g an
+gang	g ang
+gao	g ao
+ge	g e
+gei	g ei
+gen	g en
+geng	g eng
+gong	g ong
+gou	g ou
+gu	g u
+gua	g ua
+guai	g uai
+guan	g uan
+guang	g uang
+gui	g ui
+gun	g un
+guo	g uo
+ha	h a
+hai	h ai
+han	h an
+hang	h ang
+hao	h ao
+he	h e
+hei	h ei
+hen	h en
+heng	h eng
+hong	h ong
+hou	h ou
+hu	h u
+hua	h ua
+huai	h uai
+huan	h uan
+huang	h uang
+hui	h ui
+hun	h un
+huo	h uo
+ji	j i
+jia	j ia
+jian	j ian
+jiang	j iang
+jiao	j iao
+jie	j ie
+jin	j in
+jing	j ing
+jiong	j iong
+jiu	j iu
+ju	j v
+jv	j v
+juan	j van
+jvan	j van
+jue	j ve
+jve	j ve
+jun	j vn
+jvn	j vn
+ka	k a
+kai	k ai
+kan	k an
+kang	k ang
+kao	k ao
+ke	k e
+kei	k ei
+ken	k en
+keng	k eng
+kong	k ong
+kou	k ou
+ku	k u
+kua	k ua
+kuai	k uai
+kuan	k uan
+kuang	k uang
+kui	k ui
+kun	k un
+kuo	k uo
+la	l a
+lai	l ai
+lan	l an
+lang	l ang
+lao	l ao
+le	l e
+lei	l ei
+leng	l eng
+li	l i
+lia	l ia
+lian	l ian
+liang	l iang
+liao	l iao
+lie	l ie
+lin	l in
+ling	l ing
+liu	l iu
+lo	l o
+long	l ong
+lou	l ou
+lu	l u
+luan	l uan
+lun	l un
+luo	l uo
+lv	l v
+lve	l ve
+ma	m a
+mai	m ai
+man	m an
+mang	m ang
+mao	m ao
+me	m e
+mei	m ei
+men	m en
+meng	m eng
+mi	m i
+mian	m ian
+miao	m iao
+mie	m ie
+min	m in
+ming	m ing
+miu	m iu
+mo	m o
+mou	m ou
+mu	m u
+na	n a
+nai	n ai
+nan	n an
+nang	n ang
+nao	n ao
+ne	n e
+nei	n ei
+nen	n en
+neng	n eng
+ni	n i
+nian	n ian
+niang	n iang
+niao	n iao
+nie	n ie
+nin	n in
+ning	n ing
+niu	n iu
+nong	n ong
+nou	n ou
+nu	n u
+nuan	n uan
+nun	n un
+nuo	n uo
+nv	n v
+nve	n ve
+o	OO o
+ou	OO ou
+pa	p a
+pai	p ai
+pan	p an
+pang	p ang
+pao	p ao
+pei	p ei
+pen	p en
+peng	p eng
+pi	p i
+pian	p ian
+piao	p iao
+pie	p ie
+pin	p in
+ping	p ing
+po	p o
+pou	p ou
+pu	p u
+qi	q i
+qia	q ia
+qian	q ian
+qiang	q iang
+qiao	q iao
+qie	q ie
+qin	q in
+qing	q ing
+qiong	q iong
+qiu	q iu
+qu	q v
+qv	q v
+quan	q van
+qvan	q van
+que	q ve
+qve	q ve
+qun	q vn
+qvn	q vn
+ran	r an
+rang	r ang
+rao	r ao
+re	r e
+ren	r en
+reng	r eng
+ri	r ir
+rong	r ong
+rou	r ou
+ru	r u
+rua	r ua
+ruan	r uan
+rui	r ui
+run	r un
+ruo	r uo
+sa	s a
+sai	s ai
+san	s an
+sang	s ang
+sao	s ao
+se	s e
+sen	s en
+seng	s eng
+sha	sh a
+shai	sh ai
+shan	sh an
+shang	sh ang
+shao	sh ao
+she	sh e
+shei	sh ei
+shen	sh en
+sheng	sh eng
+shi	sh ir
+shou	sh ou
+shu	sh u
+shua	sh ua
+shuai	sh uai
+shuan	sh uan
+shuang	sh uang
+shui	sh ui
+shun	sh un
+shuo	sh uo
+si	s i0
+song	s ong
+sou	s ou
+su	s u
+suan	s uan
+sui	s ui
+sun	s un
+suo	s uo
+ta	t a
+tai	t ai
+tan	t an
+tang	t ang
+tao	t ao
+te	t e
+tei	t ei
+teng	t eng
+ti	t i
+tian	t ian
+tiao	t iao
+tie	t ie
+ting	t ing
+tong	t ong
+tou	t ou
+tu	t u
+tuan	t uan
+tui	t ui
+tun	t un
+tuo	t uo
+wa	w a
+wai	w ai
+wan	w an
+wang	w ang
+wei	w ei
+wen	w en
+weng	w eng
+wo	w o
+wu	w u
+xi	x i
+xia	x ia
+xian	x ian
+xiang	x iang
+xiao	x iao
+xie	x ie
+xin	x in
+xing	x ing
+xiong	x iong
+xiu	x iu
+xu	x v
+xv	x v
+xuan	x van
+xvan	x van
+xue	x ve
+xve	x ve
+xun	x vn
+xvn	x vn
+ya	y a
+yan	y En
+yang	y ang
+yao	y ao
+ye	y E
+yi	y i
+yin	y in
+ying	y ing
+yo	y o
+yong	y ong
+you	y ou
+yu	y v
+yv	y v
+yuan	y van
+yvan	y van
+yue	y ve
+yve	y ve
+yun	y vn
+yvn	y vn
+za	z a
+zai	z ai
+zan	z an
+zang	z ang
+zao	z ao
+ze	z e
+zei	z ei
+zen	z en
+zeng	z eng
+zha	zh a
+zhai	zh ai
+zhan	zh an
+zhang	zh ang
+zhao	zh ao
+zhe	zh e
+zhei	zh ei
+zhen	zh en
+zheng	zh eng
+zhi	zh ir
+zhong	zh ong
+zhou	zh ou
+zhu	zh u
+zhua	zh ua
+zhuai	zh uai
+zhuan	zh uan
+zhuang	zh uang
+zhui	zh ui
+zhun	zh un
+zhuo	zh uo
+zi	z i0
+zong	z ong
+zou	z ou
+zu	z u
+zuan	z uan
+zui	z ui
+zun	z un
+zuo	z uo

text/shanghainese.py

@@ -0,0 +1,139 @@
+import re
+
+import cn2an
+import opencc
+
+from text.symbols import punctuation, sh_symbols
+
+converter = opencc.OpenCC('text/lexicon/zaonhe.json')
+
+
+def number_to_shanghainese(text):
+    def to_shanghainese(num):
+        num = cn2an.an2cn(num).replace('一十', '十').replace('二十', '廿').replace('二', '两')
+        return re.sub(r'((?:^|[^三四五六七八九])十|廿)两', r'\1二', num)
+
+    return re.sub(r'\d+(?:\.?\d+)?', lambda x: to_shanghainese(x.group()), text)
+
+
+rep_map = {
+    "：": ",",
+    "；": ",",
+    "，": ",",
+    "。": ".",
+    "！": "!",
+    "？": "?",
+    "\n": ".",
+    "·": ",",
+    "、": ",",
+    "...": "…",
+    "$": ".",
+    "“": "'",
+    "”": "'",
+    "‘": "'",
+    "’": "'",
+    "（": "'",
+    "）": "'",
+    "(": "'",
+    ")": "'",
+    "《": "'",
+    "》": "'",
+    "【": "'",
+    "】": "'",
+    "[": "'",
+    "]": "'",
+    "—": "-",
+    "～": "-",
+    "~": "-",
+    "「": "'",
+    "」": "'",
+}
+
+
+def replace_punctuation(text):
+    text = text.replace("嗯", "恩").replace("呣", "母")
+    pattern = re.compile("|".join(re.escape(p) for p in rep_map.keys()))
+
+    replaced_text = pattern.sub(lambda x: rep_map[x.group()], text)
+
+    replaced_text = re.sub(r"[^\u4e00-\u9fa5" + "".join(punctuation) + r"]+", "", replaced_text)
+
+    return replaced_text
+
+
+def valid_tone_char(char):
+    return ("SH_" + char) in sh_symbols \
+        or char in punctuation \
+        or char.isdigit() \
+        or char.isspace()
+
+
+def g2p(text):
+    phones, tones, word2ph = _g2p(text)
+    phones = ["_"] + phones + ["_"]
+    tones = [0] + tones + [0]
+    word2ph = [1] + word2ph + [1]
+    return phones, tones, word2ph
+
+
+def _g2p(text):
+    phones = converter.convert(text).replace('-', '').replace('$', '')
+    phones = "".join([i if valid_tone_char(i) else '' for i in phones])
+    phone_chars = [i for i in phones]
+
+    phones = []
+    tones = []
+    word2ph = []
+
+    if len(phone_chars) == 0:
+        return phones, tones, word2ph
+
+    phone_start_pos = 0
+
+    for pos in range(len(phone_chars)):
+        char = phone_chars[pos]
+        if char.isdigit():
+            tone = int(char)
+            word2ph = word2ph + [pos - phone_start_pos]
+            for j in range(phone_start_pos, pos):
+                tones = tones + [tone]
+            phone_start_pos = pos + 1
+
+        elif char in punctuation:
+            if pos != phone_start_pos:
+                word2ph = word2ph + [pos - phone_start_pos]
+                for j in range(phone_start_pos, pos):
+                    tones = tones + [0]
+                pass
+            phones = phones + [char]
+            tones = tones + [0]
+            word2ph = word2ph + [1]
+            phone_start_pos = pos + 1
+
+        else:
+            phones = phones + [char]
+        pass
+
+    last_phone_char = phone_chars[-1]
+    if not last_phone_char.isdigit() and last_phone_char not in punctuation:
+        word2ph = word2ph + [len(phone_chars) - phone_start_pos]
+        for j in range(phone_start_pos, len(phone_chars)):
+            tones = tones + [0]
+        pass
+
+    # phones 加前缀 'SH'
+    phones = ["SH_" + i if ("SH_" + i) in sh_symbols else i for i in phones]
+    assert len(tones) == len(phones)
+    assert sum(word2ph) == len(phones)
+    return phones, tones, word2ph
+
+
+def text_normalize(text):
+    text = number_to_shanghainese(text.upper())
+    text = replace_punctuation(text)
+    return text
+
+
+def get_bert_feature(text, word2ph, device):
+    from text import shanghainese_bert
+    return shanghainese_bert.get_bert_feature(text, word2ph, device)

text/shanghainese_bert.py

@@ -0,0 +1,39 @@
+import sys
+
+import torch
+from transformers import AutoTokenizer, AutoModelForMaskedLM
+
+model_name = "./bert/chinese-roberta-wwm-ext-large"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+models = dict()
+
+
+def get_bert_feature(text, word2ph, device="cuda"):
+    if (
+            sys.platform == "darwin"
+            and torch.backends.mps.is_available()
+            and device == "cpu"
+    ):
+        device = "mps"
+    if not device:
+        device = "cuda"
+    if device not in models.keys():
+        models[device] = AutoModelForMaskedLM.from_pretrained(model_name).to(device)
+
+    with torch.no_grad():
+        inputs = tokenizer(text, return_tensors="pt")
+        for i in inputs:
+            inputs[i] = inputs[i].to(device)
+        res = models[device](**inputs, output_hidden_states=True)
+        res = torch.cat(res["hidden_states"][-3:-2], -1)[0].cpu()
+
+    word2phone = word2ph
+    phone_level_feature = []
+    for i in range(len(word2phone)):
+        repeat_feature = res[i].repeat(word2phone[i], 1)
+        phone_level_feature.append(repeat_feature)
+
+    phone_level_feature = torch.cat(phone_level_feature, dim=0)
+
+    return phone_level_feature.T

text/symbols.py

@@ -0,0 +1,52 @@
+punctuation = ["!", "?", "…", ",", ".", "'", "-"]
+pu_symbols = punctuation + ["SP", "UNK"]
+pad = "_"
+
+# chinese
+zh_symbols = [
+    "E", "En", "a", "ai", "an", "ang", "ao", "b", "c", "ch", "d", "e", "ei", "en", "eng", "er", "f", "g", "h", "i", "i0", "ia", "ian", "iang", "iao", "ie",
+    "in", "ing", "iong", "ir", "iu", "j", "k", "l", "m", "n", "o", "ong", "ou", "p", "q", "r", "s", "sh", "t", "u", "ua", "uai", "uan", "uang", "ui", "un",
+    "uo", "v", "van", "ve", "vn", "w", "x", "y", "z", "zh", "AA", "EE", "OO",
+]
+zh_symbols = ["ZH_" + i for i in zh_symbols]
+num_zh_tones = 6
+
+# shanghainese
+sh_symbols = [
+    "\u2026", "a", "b", "d", "f", "g", "h", "i", "k", "l", "m", "n", "o", "p", "s", "t", "u", "v", "y", "z", "\u00f8", "\u014b", "\u0235", "\u0251", "\u0254",
+    "\u0255", "\u0259", "\u0264", "\u0266", "\u026a", "\u027f", "\u0291", "\u0294", "\u02b0", "\u0303", "\u0329", "\u1d00", "\u1d07"
+]
+sh_symbols = ["SH_" + i for i in sh_symbols]
+num_sh_tones = 9
+
+# English
+en_symbols = [
+    "aa", "ae", "ah", "ao", "aw", "ay", "b", "ch", "d", "dh", "eh", "er", "ey", "f", "g", "hh", "ih", "iy", "jh", "k", "l", "m", "n", "ng", "ow", "oy", "p",
+    "r", "s", "sh", "t", "th", "uh", "uw", "V", "w", "y", "z", "zh",
+]
+en_symbols = ["EN_" + i for i in en_symbols]
+num_en_tones = 4
+
+# combine all symbols
+normal_symbols = sorted(set(zh_symbols + sh_symbols + en_symbols))
+symbols = [pad] + normal_symbols + pu_symbols
+sil_phonemes_ids = [symbols.index(i) for i in pu_symbols]
+
+# combine all tones
+num_tones = num_zh_tones + num_sh_tones + num_en_tones
+
+# language maps
+language_id_map = {"ZH": 0, "SH": 1, "EN": 2}
+num_languages = len(language_id_map.keys())
+
+language_tone_start_map = {
+    "ZH": 0,
+    "SH": num_zh_tones,
+    "EN": num_zh_tones + num_sh_tones,
+}
+
+if __name__ == "__main__":
+    a = set(zh_symbols)
+    b = set(sh_symbols)
+    c = set(en_symbols)
+    print(sorted(a & b & c))

text/tone_sandhi.py

@@ -0,0 +1,769 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from typing import List
+from typing import Tuple
+
+import jieba
+from pypinyin import lazy_pinyin
+from pypinyin import Style
+
+
+class ToneSandhi:
+    def __init__(self):
+        self.must_neural_tone_words = {
+            "麻烦",
+            "麻利",
+            "鸳鸯",
+            "高粱",
+            "骨头",
+            "骆驼",
+            "马虎",
+            "首饰",
+            "馒头",
+            "馄饨",
+            "风筝",
+            "难为",
+            "队伍",
+            "阔气",
+            "闺女",
+            "门道",
+            "锄头",
+            "铺盖",
+            "铃铛",
+            "铁匠",
+            "钥匙",
+            "里脊",
+            "里头",
+            "部分",
+            "那么",
+            "道士",
+            "造化",
+            "迷糊",
+            "连累",
+            "这么",
+            "这个",
+            "运气",
+            "过去",
+            "软和",
+            "转悠",
+            "踏实",
+            "跳蚤",
+            "跟头",
+            "趔趄",
+            "财主",
+            "豆腐",
+            "讲究",
+            "记性",
+            "记号",
+            "认识",
+            "规矩",
+            "见识",
+            "裁缝",
+            "补丁",
+            "衣裳",
+            "衣服",
+            "衙门",
+            "街坊",
+            "行李",
+            "行当",
+            "蛤蟆",
+            "蘑菇",
+            "薄荷",
+            "葫芦",
+            "葡萄",
+            "萝卜",
+            "荸荠",
+            "苗条",
+            "苗头",
+            "苍蝇",
+            "芝麻",
+            "舒服",
+            "舒坦",
+            "舌头",
+            "自在",
+            "膏药",
+            "脾气",
+            "脑袋",
+            "脊梁",
+            "能耐",
+            "胳膊",
+            "胭脂",
+            "胡萝",
+            "胡琴",
+            "胡同",
+            "聪明",
+            "耽误",
+            "耽搁",
+            "耷拉",
+            "耳朵",
+            "老爷",
+            "老实",
+            "老婆",
+            "老头",
+            "老太",
+            "翻腾",
+            "罗嗦",
+            "罐头",
+            "编辑",
+            "结实",
+            "红火",
+            "累赘",
+            "糨糊",
+            "糊涂",
+            "精神",
+            "粮食",
+            "簸箕",
+            "篱笆",
+            "算计",
+            "算盘",
+            "答应",
+            "笤帚",
+            "笑语",
+            "笑话",
+            "窟窿",
+            "窝囊",
+            "窗户",
+            "稳当",
+            "稀罕",
+            "称呼",
+            "秧歌",
+            "秀气",
+            "秀才",
+            "福气",
+            "祖宗",
+            "砚台",
+            "码头",
+            "石榴",
+            "石头",
+            "石匠",
+            "知识",
+            "眼睛",
+            "眯缝",
+            "眨巴",
+            "眉毛",
+            "相声",
+            "盘算",
+            "白净",
+            "痢疾",
+            "痛快",
+            "疟疾",
+            "疙瘩",
+            "疏忽",
+            "畜生",
+            "生意",
+            "甘蔗",
+            "琵琶",
+            "琢磨",
+            "琉璃",
+            "玻璃",
+            "玫瑰",
+            "玄乎",
+            "狐狸",
+            "状元",
+            "特务",
+            "牲口",
+            "牙碜",
+            "牌楼",
+            "爽快",
+            "爱人",
+            "热闹",
+            "烧饼",
+            "烟筒",
+            "烂糊",
+            "点心",
+            "炊帚",
+            "灯笼",
+            "火候",
+            "漂亮",
+            "滑溜",
+            "溜达",
+            "温和",
+            "清楚",
+            "消息",
+            "浪头",
+            "活泼",
+            "比方",
+            "正经",
+            "欺负",
+            "模糊",
+            "槟榔",
+            "棺材",
+            "棒槌",
+            "棉花",
+            "核桃",
+            "栅栏",
+            "柴火",
+            "架势",
+            "枕头",
+            "枇杷",
+            "机灵",
+            "本事",
+            "木头",
+            "木匠",
+            "朋友",
+            "月饼",
+            "月亮",
+            "暖和",
+            "明白",
+            "时候",
+            "新鲜",
+            "故事",
+            "收拾",
+            "收成",
+            "提防",
+            "挖苦",
+            "挑剔",
+            "指甲",
+            "指头",
+            "拾掇",
+            "拳头",
+            "拨弄",
+            "招牌",
+            "招呼",
+            "抬举",
+            "护士",
+            "折腾",
+            "扫帚",
+            "打量",
+            "打算",
+            "打点",
+            "打扮",
+            "打听",
+            "打发",
+            "扎实",
+            "扁担",
+            "戒指",
+            "懒得",
+            "意识",
+            "意思",
+            "情形",
+            "悟性",
+            "怪物",
+            "思量",
+            "怎么",
+            "念头",
+            "念叨",
+            "快活",
+            "忙活",
+            "志气",
+            "心思",
+            "得罪",
+            "张罗",
+            "弟兄",
+            "开通",
+            "应酬",
+            "庄稼",
+            "干事",
+            "帮手",
+            "帐篷",
+            "希罕",
+            "师父",
+            "师傅",
+            "巴结",
+            "巴掌",
+            "差事",
+            "工夫",
+            "岁数",
+            "屁股",
+            "尾巴",
+            "少爷",
+            "小气",
+            "小伙",
+            "将就",
+            "对头",
+            "对付",
+            "寡妇",
+            "家伙",
+            "客气",
+            "实在",
+            "官司",
+            "学问",
+            "学生",
+            "字号",
+            "嫁妆",
+            "媳妇",
+            "媒人",
+            "婆家",
+            "娘家",
+            "委屈",
+            "姑娘",
+            "姐夫",
+            "妯娌",
+            "妥当",
+            "妖精",
+            "奴才",
+            "女婿",
+            "头发",
+            "太阳",
+            "大爷",
+            "大方",
+            "大意",
+            "大夫",
+            "多少",
+            "多么",
+            "外甥",
+            "壮实",
+            "地道",
+            "地方",
+            "在乎",
+            "困难",
+            "嘴巴",
+            "嘱咐",
+            "嘟囔",
+            "嘀咕",
+            "喜欢",
+            "喇嘛",
+            "喇叭",
+            "商量",
+            "唾沫",
+            "哑巴",
+            "哈欠",
+            "哆嗦",
+            "咳嗽",
+            "和尚",
+            "告诉",
+            "告示",
+            "含糊",
+            "吓唬",
+            "后头",
+            "名字",
+            "名堂",
+            "合同",
+            "吆喝",
+            "叫唤",
+            "口袋",
+            "厚道",
+            "厉害",
+            "千斤",
+            "包袱",
+            "包涵",
+            "匀称",
+            "勤快",
+            "动静",
+            "动弹",
+            "功夫",
+            "力气",
+            "前头",
+            "刺猬",
+            "刺激",
+            "别扭",
+            "利落",
+            "利索",
+            "利害",
+            "分析",
+            "出息",
+            "凑合",
+            "凉快",
+            "冷战",
+            "冤枉",
+            "冒失",
+            "养活",
+            "关系",
+            "先生",
+            "兄弟",
+            "便宜",
+            "使唤",
+            "佩服",
+            "作坊",
+            "体面",
+            "位置",
+            "似的",
+            "伙计",
+            "休息",
+            "什么",
+            "人家",
+            "亲戚",
+            "亲家",
+            "交情",
+            "云彩",
+            "事情",
+            "买卖",
+            "主意",
+            "丫头",
+            "丧气",
+            "两口",
+            "东西",
+            "东家",
+            "世故",
+            "不由",
+            "不在",
+            "下水",
+            "下巴",
+            "上头",
+            "上司",
+            "丈夫",
+            "丈人",
+            "一辈",
+            "那个",
+            "菩萨",
+            "父亲",
+            "母亲",
+            "咕噜",
+            "邋遢",
+            "费用",
+            "冤家",
+            "甜头",
+            "介绍",
+            "荒唐",
+            "大人",
+            "泥鳅",
+            "幸福",
+            "熟悉",
+            "计划",
+            "扑腾",
+            "蜡烛",
+            "姥爷",
+            "照顾",
+            "喉咙",
+            "吉他",
+            "弄堂",
+            "蚂蚱",
+            "凤凰",
+            "拖沓",
+            "寒碜",
+            "糟蹋",
+            "倒腾",
+            "报复",
+            "逻辑",
+            "盘缠",
+            "喽啰",
+            "牢骚",
+            "咖喱",
+            "扫把",
+            "惦记",
+        }
+        self.must_not_neural_tone_words = {
+            "男子",
+            "女子",
+            "分子",
+            "原子",
+            "量子",
+            "莲子",
+            "石子",
+            "瓜子",
+            "电子",
+            "人人",
+            "虎虎",
+        }
+        self.punc = "：，；。？！“”‘’':,;.?!"
+
+    # the meaning of jieba pos tag: https://blog.csdn.net/weixin_44174352/article/details/113731041
+    # e.g.
+    # word: "家里"
+    # pos: "s"
+    # finals: ['ia1', 'i3']
+    def _neural_sandhi(self, word: str, pos: str, finals: List[str]) -> List[str]:
+        # reduplication words for n. and v. e.g. 奶奶, 试试, 旺旺
+        for j, item in enumerate(word):
+            if (
+                j - 1 >= 0
+                and item == word[j - 1]
+                and pos[0] in {"n", "v", "a"}
+                and word not in self.must_not_neural_tone_words
+            ):
+                finals[j] = finals[j][:-1] + "5"
+        ge_idx = word.find("个")
+        if len(word) >= 1 and word[-1] in "吧呢啊呐噻嘛吖嗨呐哦哒额滴哩哟喽啰耶喔诶":
+            finals[-1] = finals[-1][:-1] + "5"
+        elif len(word) >= 1 and word[-1] in "的地得":
+            finals[-1] = finals[-1][:-1] + "5"
+        # e.g. 走了, 看着, 去过
+        # elif len(word) == 1 and word in "了着过" and pos in {"ul", "uz", "ug"}:
+        #     finals[-1] = finals[-1][:-1] + "5"
+        elif (
+            len(word) > 1
+            and word[-1] in "们子"
+            and pos in {"r", "n"}
+            and word not in self.must_not_neural_tone_words
+        ):
+            finals[-1] = finals[-1][:-1] + "5"
+        # e.g. 桌上, 地下, 家里
+        elif len(word) > 1 and word[-1] in "上下里" and pos in {"s", "l", "f"}:
+            finals[-1] = finals[-1][:-1] + "5"
+        # e.g. 上来, 下去
+        elif len(word) > 1 and word[-1] in "来去" and word[-2] in "上下进出回过起开":
+            finals[-1] = finals[-1][:-1] + "5"
+        # 个做量词
+        elif (
+            ge_idx >= 1
+            and (word[ge_idx - 1].isnumeric() or word[ge_idx - 1] in "几有两半多各整每做是")
+        ) or word == "个":
+            finals[ge_idx] = finals[ge_idx][:-1] + "5"
+        else:
+            if (
+                word in self.must_neural_tone_words
+                or word[-2:] in self.must_neural_tone_words
+            ):
+                finals[-1] = finals[-1][:-1] + "5"
+
+        word_list = self._split_word(word)
+        finals_list = [finals[: len(word_list[0])], finals[len(word_list[0]) :]]
+        for i, word in enumerate(word_list):
+            # conventional neural in Chinese
+            if (
+                word in self.must_neural_tone_words
+                or word[-2:] in self.must_neural_tone_words
+            ):
+                finals_list[i][-1] = finals_list[i][-1][:-1] + "5"
+        finals = sum(finals_list, [])
+        return finals
+
+    def _bu_sandhi(self, word: str, finals: List[str]) -> List[str]:
+        # e.g. 看不懂
+        if len(word) == 3 and word[1] == "不":
+            finals[1] = finals[1][:-1] + "5"
+        else:
+            for i, char in enumerate(word):
+                # "不" before tone4 should be bu2, e.g. 不怕
+                if char == "不" and i + 1 < len(word) and finals[i + 1][-1] == "4":
+                    finals[i] = finals[i][:-1] + "2"
+        return finals
+
+    def _yi_sandhi(self, word: str, finals: List[str]) -> List[str]:
+        # "一" in number sequences, e.g. 一零零, 二一零
+        if word.find("一") != -1 and all(
+            [item.isnumeric() for item in word if item != "一"]
+        ):
+            return finals
+        # "一" between reduplication words should be yi5, e.g. 看一看
+        elif len(word) == 3 and word[1] == "一" and word[0] == word[-1]:
+            finals[1] = finals[1][:-1] + "5"
+        # when "一" is ordinal word, it should be yi1
+        elif word.startswith("第一"):
+            finals[1] = finals[1][:-1] + "1"
+        else:
+            for i, char in enumerate(word):
+                if char == "一" and i + 1 < len(word):
+                    # "一" before tone4 should be yi2, e.g. 一段
+                    if finals[i + 1][-1] == "4":
+                        finals[i] = finals[i][:-1] + "2"
+                    # "一" before non-tone4 should be yi4, e.g. 一天
+                    else:
+                        # "一" 后面如果是标点，还读一声
+                        if word[i + 1] not in self.punc:
+                            finals[i] = finals[i][:-1] + "4"
+        return finals
+
+    def _split_word(self, word: str) -> List[str]:
+        word_list = jieba.cut_for_search(word)
+        word_list = sorted(word_list, key=lambda i: len(i), reverse=False)
+        first_subword = word_list[0]
+        first_begin_idx = word.find(first_subword)
+        if first_begin_idx == 0:
+            second_subword = word[len(first_subword) :]
+            new_word_list = [first_subword, second_subword]
+        else:
+            second_subword = word[: -len(first_subword)]
+            new_word_list = [second_subword, first_subword]
+        return new_word_list
+
+    def _three_sandhi(self, word: str, finals: List[str]) -> List[str]:
+        if len(word) == 2 and self._all_tone_three(finals):
+            finals[0] = finals[0][:-1] + "2"
+        elif len(word) == 3:
+            word_list = self._split_word(word)
+            if self._all_tone_three(finals):
+                #  disyllabic + monosyllabic, e.g. 蒙古/包
+                if len(word_list[0]) == 2:
+                    finals[0] = finals[0][:-1] + "2"
+                    finals[1] = finals[1][:-1] + "2"
+                #  monosyllabic + disyllabic, e.g. 纸/老虎
+                elif len(word_list[0]) == 1:
+                    finals[1] = finals[1][:-1] + "2"
+            else:
+                finals_list = [finals[: len(word_list[0])], finals[len(word_list[0]) :]]
+                if len(finals_list) == 2:
+                    for i, sub in enumerate(finals_list):
+                        # e.g. 所有/人
+                        if self._all_tone_three(sub) and len(sub) == 2:
+                            finals_list[i][0] = finals_list[i][0][:-1] + "2"
+                        # e.g. 好/喜欢
+                        elif (
+                            i == 1
+                            and not self._all_tone_three(sub)
+                            and finals_list[i][0][-1] == "3"
+                            and finals_list[0][-1][-1] == "3"
+                        ):
+                            finals_list[0][-1] = finals_list[0][-1][:-1] + "2"
+                        finals = sum(finals_list, [])
+        # split idiom into two words who's length is 2
+        elif len(word) == 4:
+            finals_list = [finals[:2], finals[2:]]
+            finals = []
+            for sub in finals_list:
+                if self._all_tone_three(sub):
+                    sub[0] = sub[0][:-1] + "2"
+                finals += sub
+
+        return finals
+
+    def _all_tone_three(self, finals: List[str]) -> bool:
+        return all(x[-1] == "3" for x in finals)
+
+    # merge "不" and the word behind it
+    # if don't merge, "不" sometimes appears alone according to jieba, which may occur sandhi error
+    def _merge_bu(self, seg: List[Tuple[str, str]]) -> List[Tuple[str, str]]:
+        new_seg = []
+        last_word = ""
+        for word, pos in seg:
+            if last_word == "不":
+                word = last_word + word
+            if word != "不":
+                new_seg.append((word, pos))
+            last_word = word[:]
+        if last_word == "不":
+            new_seg.append((last_word, "d"))
+            last_word = ""
+        return new_seg
+
+    # function 1: merge "一" and reduplication words in it's left and right, e.g. "听","一","听" ->"听一听"
+    # function 2: merge single  "一" and the word behind it
+    # if don't merge, "一" sometimes appears alone according to jieba, which may occur sandhi error
+    # e.g.
+    # input seg: [('听', 'v'), ('一', 'm'), ('听', 'v')]
+    # output seg: [['听一听', 'v']]
+    def _merge_yi(self, seg: List[Tuple[str, str]]) -> List[Tuple[str, str]]:
+        new_seg = []
+        # function 1
+        for i, (word, pos) in enumerate(seg):
+            if (
+                i - 1 >= 0
+                and word == "一"
+                and i + 1 < len(seg)
+                and seg[i - 1][0] == seg[i + 1][0]
+                and seg[i - 1][1] == "v"
+            ):
+                new_seg[i - 1][0] = new_seg[i - 1][0] + "一" + new_seg[i - 1][0]
+            else:
+                if (
+                    i - 2 >= 0
+                    and seg[i - 1][0] == "一"
+                    and seg[i - 2][0] == word
+                    and pos == "v"
+                ):
+                    continue
+                else:
+                    new_seg.append([word, pos])
+        seg = new_seg
+        new_seg = []
+        # function 2
+        for i, (word, pos) in enumerate(seg):
+            if new_seg and new_seg[-1][0] == "一":
+                new_seg[-1][0] = new_seg[-1][0] + word
+            else:
+                new_seg.append([word, pos])
+        return new_seg
+
+    # the first and the second words are all_tone_three
+    def _merge_continuous_three_tones(
+        self, seg: List[Tuple[str, str]]
+    ) -> List[Tuple[str, str]]:
+        new_seg = []
+        sub_finals_list = [
+            lazy_pinyin(word, neutral_tone_with_five=True, style=Style.FINALS_TONE3)
+            for (word, pos) in seg
+        ]
+        assert len(sub_finals_list) == len(seg)
+        merge_last = [False] * len(seg)
+        for i, (word, pos) in enumerate(seg):
+            if (
+                i - 1 >= 0
+                and self._all_tone_three(sub_finals_list[i - 1])
+                and self._all_tone_three(sub_finals_list[i])
+                and not merge_last[i - 1]
+            ):
+                # if the last word is reduplication, not merge, because reduplication need to be _neural_sandhi
+                if (
+                    not self._is_reduplication(seg[i - 1][0])
+                    and len(seg[i - 1][0]) + len(seg[i][0]) <= 3
+                ):
+                    new_seg[-1][0] = new_seg[-1][0] + seg[i][0]
+                    merge_last[i] = True
+                else:
+                    new_seg.append([word, pos])
+            else:
+                new_seg.append([word, pos])
+
+        return new_seg
+
+    def _is_reduplication(self, word: str) -> bool:
+        return len(word) == 2 and word[0] == word[1]
+
+    # the last char of first word and the first char of second word is tone_three
+    def _merge_continuous_three_tones_2(
+        self, seg: List[Tuple[str, str]]
+    ) -> List[Tuple[str, str]]:
+        new_seg = []
+        sub_finals_list = [
+            lazy_pinyin(word, neutral_tone_with_five=True, style=Style.FINALS_TONE3)
+            for (word, pos) in seg
+        ]
+        assert len(sub_finals_list) == len(seg)
+        merge_last = [False] * len(seg)
+        for i, (word, pos) in enumerate(seg):
+            if (
+                i - 1 >= 0
+                and sub_finals_list[i - 1][-1][-1] == "3"
+                and sub_finals_list[i][0][-1] == "3"
+                and not merge_last[i - 1]
+            ):
+                # if the last word is reduplication, not merge, because reduplication need to be _neural_sandhi
+                if (
+                    not self._is_reduplication(seg[i - 1][0])
+                    and len(seg[i - 1][0]) + len(seg[i][0]) <= 3
+                ):
+                    new_seg[-1][0] = new_seg[-1][0] + seg[i][0]
+                    merge_last[i] = True
+                else:
+                    new_seg.append([word, pos])
+            else:
+                new_seg.append([word, pos])
+        return new_seg
+
+    def _merge_er(self, seg: List[Tuple[str, str]]) -> List[Tuple[str, str]]:
+        new_seg = []
+        for i, (word, pos) in enumerate(seg):
+            if i - 1 >= 0 and word == "儿" and seg[i - 1][0] != "#":
+                new_seg[-1][0] = new_seg[-1][0] + seg[i][0]
+            else:
+                new_seg.append([word, pos])
+        return new_seg
+
+    def _merge_reduplication(self, seg: List[Tuple[str, str]]) -> List[Tuple[str, str]]:
+        new_seg = []
+        for i, (word, pos) in enumerate(seg):
+            if new_seg and word == new_seg[-1][0]:
+                new_seg[-1][0] = new_seg[-1][0] + seg[i][0]
+            else:
+                new_seg.append([word, pos])
+        return new_seg
+
+    def pre_merge_for_modify(self, seg: List[Tuple[str, str]]) -> List[Tuple[str, str]]:
+        seg = self._merge_bu(seg)
+        try:
+            seg = self._merge_yi(seg)
+        except:
+            print("_merge_yi failed")
+        seg = self._merge_reduplication(seg)
+        seg = self._merge_continuous_three_tones(seg)
+        seg = self._merge_continuous_three_tones_2(seg)
+        seg = self._merge_er(seg)
+        return seg
+
+    def modified_tone(self, word: str, pos: str, finals: List[str]) -> List[str]:
+        finals = self._bu_sandhi(word, finals)
+        finals = self._yi_sandhi(word, finals)
+        finals = self._neural_sandhi(word, pos, finals)
+        finals = self._three_sandhi(word, finals)
+        return finals