Cloyir/index-tts2-ForDgxSpark
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

一些输入文本处理的fix (#18)

Browse Source 
* +简单前端

* 前端兼容arm机器

* fix

* fix

* WeTextProcessing: overwrite_cache=True 刷新前端缓存

* 恢复输入中的拼音

---------
Co-authored-by: kemuriririn <10inspiral@gmail.com>
This commit is contained in:
kemuriririn 2025-03-27 14:12:12 +08:00 committed by GitHub
parent c73344ecc9
commit 1734698f3d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 78 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
indextts/infer.py
View File

@ -58,7 +58,10 @@ class IndexTTS:
return self.normalizer.infer(text)
def infer(self, audio_prompt, text, output_path):
print(f"origin text:{text}")
text = self.preprocess_text(text)
print(f"normalized text:{text}")
audio, sr = torchaudio.load(audio_prompt)
audio = torch.mean(audio, dim=0, keepdim=True)

85
indextts/utils/front.py
View File

@ -1,12 +1,6 @@
# -*- coding: utf-8 -*-
import traceback
import os
import sys
import re
import re
class TextNormalizer:
def __init__(self):
@ -69,16 +63,15 @@ class TextNormalizer:
# print(os.path.join(os.path.dirname(os.path.abspath(__file__)), ".."))
# sys.path.append(model_dir)
import platform
if platform.machine() == "aarch64":
if platform.system() == "Darwin":
from wetext import Normalizer
self.zh_normalizer = Normalizer(remove_erhua=False,lang="zh",operator="tn")
self.en_normalizer = Normalizer(lang="en",operator="tn")
else:
from tn.chinese.normalizer import Normalizer as NormalizerZh
from tn.english.normalizer import Normalizer as NormalizerEn
self.zh_normalizer = NormalizerZh(remove_interjections=False, remove_erhua=False)
self.en_normalizer = NormalizerEn()
self.zh_normalizer = NormalizerZh(remove_interjections=False, remove_erhua=False,overwrite_cache=True)
self.en_normalizer = NormalizerEn(overwrite_cache=True)
def infer(self, text):
pattern = re.compile("|".join(re.escape(p) for p in self.char_rep_map.keys()))
@ -92,8 +85,80 @@ class TextNormalizer:
except Exception:
result = ""
print(traceback.format_exc())
result = self.restore_pinyin_tone_numbers(replaced_text, result)
return result
def pinyin_match(self, pinyin):
pattern = r"(qun)(\d)"
repl = r"qvn\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(quan)(\d)"
repl = r"qvan\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(que)(\d)"
repl = r"qve\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(qu)(\d)"
repl = r"qv\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(ju)(\d)"
repl = r"jv\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(jue)(\d)"
repl = r"jve\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(xun)(\d)"
repl = r"xvn\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(xue)(\d)"
repl = r"xve\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(xu)(\d)"
repl = r"xv\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(juan)(\d)"
repl = r"jvan\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(jun)(\d)"
repl = r"jvn\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
pattern = r"(xuan)(\d)"
repl = r"xvan\g<2>"
pinyin = re.sub(pattern, repl, pinyin)
return pinyin
def restore_pinyin_tone_numbers(self,original_text, processed_text):
# 第一步：恢复拼音后的音调数字（1-4）
# 建立中文数字到阿拉伯数字的映射
chinese_to_num = {'一': '1', '二': '2', '三': '3', '四': '4'}
# 使用正则表达式找到拼音+中文数字的组合（如 "xuan四"）
def replace_tone(match):
pinyin = match.group(1) # 拼音部分
chinese_num = match.group(2) # 中文数字部分
# 将中文数字转换为阿拉伯数字
num = chinese_to_num.get(chinese_num, chinese_num)
return f"{pinyin}{num}"
# 匹配拼音后跟中文数字（一、二、三、四）的情况
pattern = r'([a-zA-Z]+)([一二三四])'
restored_text = re.sub(pattern, replace_tone, processed_text)
restored_text = restored_text.lower()
restored_text = self.pinyin_match(restored_text)
return restored_text
if __name__ == '__main__':
# 测试程序