Spaces:
Running
Running
from text.symbols import punctuation | |
import re | |
import unicodedata | |
import cn2an | |
import pycantonese | |
import jieba | |
import csv | |
jieba.load_userdict("./text/yue_dict.txt") | |
jyutping_dict = {} | |
with open("./text/jyutping.csv", "r", encoding="utf-8") as f: | |
for line in f: | |
line = line.strip() | |
if not line: | |
continue | |
word, jyutping = line.split(",") | |
if word not in jyutping_dict: | |
jyutping_dict[word] = [jyutping] | |
else: | |
jyutping_dict[word].append(jyutping) | |
def normalizer(x): | |
x = cn2an.transform(x, "an2cn") | |
return x | |
def word2jyutping(word): | |
jyutpings = [pycantonese.characters_to_jyutping( | |
w)[0][1] for w in word if unicodedata.name(w, "").startswith("CJK UNIFIED IDEOGRAPH")] | |
for i, j in enumerate(jyutpings): | |
if re.search(r"^(la|ga)[1-6]$", j): | |
# la1 -> laa1, ga1 -> gaa1 | |
jyutpings[i] = jyutpings[i].replace('a', 'aa') | |
if None in jyutpings: | |
raise ValueError(f"Failed to convert {word} to jyutping: {jyutpings}") | |
return " ".join(jyutpings) | |
INITIALS = ["", "b", "c", "d", "f", "g", "gw", "h", "j", | |
"k", "kw", "l", "m", "n", "ng", "p", "s", "t", "w", "z"] | |
FINALS = ["aa", "aai", "aau", "aam", "aan", "aang", "aap", "aat", "aak", "ai", "au", "am", "an", "ang", "ap", "at", "ak", "e", "ei", "eu", "em", "eng", "ep", "ek", "i", "iu", "im", | |
"in", "ing", "ip", "it", "ik", "o", "oi", "ou", "on", "ong", "ot", "ok", "oe", "oeng", "oek", "eoi", "eon", "eot", "u", "ui", "un", "ung", "ut", "uk", "yu", "yun", "yut", "m", "ng"] | |
rep_map = { | |
":": ",", | |
"︰": ",", | |
";": ",", | |
",": ",", | |
"﹐": ",", | |
"。": ".", | |
"!": "!", | |
"?": "?", | |
"﹖": "?", | |
"﹗": "!", | |
"\n": ".", | |
"·": ",", | |
"、": ",", | |
"丶": ",", | |
"...": "…", | |
"⋯": "…", | |
"$": ".", | |
"“": "'", | |
"”": "'", | |
'"': "'", | |
"‘": "'", | |
"’": "'", | |
"(": "'", | |
")": "'", | |
"(": "'", | |
")": "'", | |
"《": "'", | |
"》": "'", | |
"【": "'", | |
"】": "'", | |
"[": "'", | |
"]": "'", | |
"—": "-", | |
"~": "-", | |
"~": "-", | |
"「": "'", | |
"」": "'", | |
"_": "-", | |
} | |
replacement_chars = { | |
"\n": " ", | |
'ㄧ': '一', | |
'—': '一', | |
'更': '更', | |
'不': '不', | |
'料': '料', | |
'聯': '聯', | |
'行': '行', | |
'利': '利', | |
'謢': '護', | |
'岀': '出', | |
'鎭': '鎮', | |
'戯': '戲', | |
'旣': '既', | |
'立': '立', | |
'來': '來', | |
'年': '年', | |
'㗇': '蝦', | |
} | |
def replace_punctuation(text): | |
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 = "".join( | |
c for c in replaced_text if unicodedata.name(c, "").startswith("CJK UNIFIED IDEOGRAPH") or c in punctuation | |
) | |
return replaced_text | |
def replace_chars(text): | |
for k, v in replacement_chars.items(): | |
text = text.replace(k, v) | |
return text | |
def word_segmentation(text): | |
words = jieba.cut(text) | |
return words | |
def text_normalize(text): | |
text = text.strip() | |
text = normalizer(text) | |
text = replace_punctuation(text) | |
text = replace_chars(text) | |
return text | |
def jyuping_to_initials_finals_tones(jyuping_syllables): | |
initials_finals = [] | |
tones = [] | |
word2ph = [] | |
for syllable in jyuping_syllables: | |
if syllable in punctuation: | |
initials_finals.append(syllable) | |
tones.append(0) | |
word2ph.append(1) # Add 1 for punctuation | |
else: | |
init, final, tone = parse_jyutping(syllable) | |
initials_finals.extend([init, final]) | |
tones.extend([tone, tone]) | |
word2ph.append(2) | |
assert len(initials_finals) == len(tones) | |
return initials_finals, tones, word2ph | |
wordshk_juytping = {} | |
# with open("/notebooks/bert-vits2/Bert-VITS2-Cantonese/wordshk_juytping.csv", "r") as csv_file: | |
# csv_reader = csv.reader(csv_file, delimiter=',') | |
# for row in csv_reader: | |
# wordshk_juytping[text_normalize(row[0])] = row[1] | |
def get_jyutping(text): | |
if text in wordshk_juytping: | |
return wordshk_juytping[text].split(" ") | |
words = word_segmentation(text) | |
jyutping_array = [] | |
punct_pattern = re.compile( | |
r"^[{}]+$".format(re.escape("".join(punctuation)))) | |
for word in words: | |
if punct_pattern.match(word): | |
puncts = re.split(r"([{}])".format( | |
re.escape("".join(punctuation))), word) | |
for punct in puncts: | |
if len(punct) > 0: | |
jyutping_array.append(punct) | |
else: | |
jyutpings = "" | |
if word in jyutping_dict: | |
jyutpings = jyutping_dict[word][0] | |
else: | |
jyutpings = word2jyutping(word) | |
# match multple jyutping eg: liu4 ge3, or single jyutping eg: liu4 | |
if not re.search(r"^([a-z]+[1-6]+[ ]?)+$", jyutpings): | |
raise ValueError( | |
f"Failed to convert {word} to jyutping: {jyutpings}") | |
jyutping_array.extend(jyutpings.split(" ")) | |
return jyutping_array | |
def get_bert_feature(text, word2ph): | |
from text import cantonese_bert | |
return cantonese_bert.get_bert_feature(text, word2ph) | |
def parse_jyutping(jyutping): | |
orig_jyutping = jyutping | |
if len(jyutping) < 2: | |
raise ValueError(f"Jyutping string too short: {jyutping}") | |
init = "" | |
if jyutping[0] == 'n' and jyutping[1] == 'g' and len(jyutping) == 3: | |
init = "" | |
elif jyutping[0] == 'm' and len(jyutping) == 2: | |
init = "" | |
elif jyutping[0] == 'n' and jyutping[1] == 'g': | |
init = 'ng' | |
jyutping = jyutping[2:] | |
elif jyutping[0] == 'g' and jyutping[1] == 'w': | |
init = 'gw' | |
jyutping = jyutping[2:] | |
elif jyutping[0] == 'k' and jyutping[1] == 'w': | |
init = 'kw' | |
jyutping = jyutping[2:] | |
elif jyutping[0] in 'bpmfdtnlgkhwzcsj': | |
init = jyutping[0] | |
jyutping = jyutping[1:] | |
else: | |
jyutping = jyutping | |
try: | |
tone = int(jyutping[-1]) | |
jyutping = jyutping[:-1] | |
except: | |
raise ValueError( | |
f"Jyutping string does not end with a tone number, in {orig_jyutping}") | |
final = jyutping | |
assert init in INITIALS, f"Invalid initial: {init}, in {orig_jyutping}" | |
if final not in FINALS: | |
raise ValueError(f"Invalid final: {final}, in {orig_jyutping}") | |
return [init, final, tone] | |
def g2p(text): | |
word2ph = [] | |
jyuping = get_jyutping(text) | |
phones, tones, word2ph = jyuping_to_initials_finals_tones(jyuping) | |
phones = ["_"] + phones + ["_"] | |
tones = [0] + tones + [0] | |
word2ph = [1] + word2ph + [1] | |
return phones, tones, word2ph | |
if __name__ == "__main__": | |
from text.cantonese_bert import get_bert_feature | |
# text = "Apple BB 你點解會咁柒㗎?我真係唔該晒你呀!123" | |
text = "佢邊係想辭工吖,跳下草裙舞想加人工之嘛。" | |
# text = "我個 app 嘅介紹文想由你寫,因為我唔知從一般用家角度要細緻到乜程度" | |
# text = "佢哋最叻咪就係去㗇人傷害人,得個殼咋!" | |
text = text_normalize(text) | |
print('normalized text', text) | |
phones, tones, word2ph = g2p(text) | |
print(phones, tones, word2ph) | |
bert = get_bert_feature(text, word2ph) | |
print(bert.shape) | |