add comment

744ea442 · BarryKCL · d17b2ee1 · 744ea442 · 744ea442 · d17b2ee1
8 changed file
--- a/README.md
+++ b/README.md
@@ -699,6 +699,7 @@ You are warmly welcome to submit questions in [discussions](https://github.com/P

 ## Acknowledgement

+- Many thanks to [BarryKCL](https://github.com/BarryKCL) improved TTS Chinses frontend based on [G2PW](https://github.com/GitYCC/g2pW)
 - Many thanks to [yeyupiaoling](https://github.com/yeyupiaoling)/[PPASR](https://github.com/yeyupiaoling/PPASR)/[PaddlePaddle-DeepSpeech](https://github.com/yeyupiaoling/PaddlePaddle-DeepSpeech)/[VoiceprintRecognition-PaddlePaddle](https://github.com/yeyupiaoling/VoiceprintRecognition-PaddlePaddle)/[AudioClassification-PaddlePaddle](https://github.com/yeyupiaoling/AudioClassification-PaddlePaddle) for years of attention, constructive advice and great help.
 - Many thanks to [mymagicpower](https://github.com/mymagicpower) for the Java implementation of ASR upon [short](https://github.com/mymagicpower/AIAS/tree/main/3_audio_sdks/asr_sdk) and [long](https://github.com/mymagicpower/AIAS/tree/main/3_audio_sdks/asr_long_audio_sdk) audio files.
 - Many thanks to [JiehangXie](https://github.com/JiehangXie)/[PaddleBoBo](https://github.com/JiehangXie/PaddleBoBo) for developing Virtual Uploader(VUP)/Virtual YouTuber(VTuber) with PaddleSpeech TTS function.

--- a/README_cn.md
+++ b/README_cn.md
@@ -833,6 +833,7 @@ PaddleSpeech 的 **语音合成** 主要包含三个模块：文本前端、声

 ## 致谢

+- 非常感谢 [BarryKCL](https://github.com/BarryKCL)基于[G2PW](https://github.com/GitYCC/g2pW)对TTS中文文本前端的优化。
 - 非常感谢 [yeyupiaoling](https://github.com/yeyupiaoling)/[PPASR](https://github.com/yeyupiaoling/PPASR)/[PaddlePaddle-DeepSpeech](https://github.com/yeyupiaoling/PaddlePaddle-DeepSpeech)/[VoiceprintRecognition-PaddlePaddle](https://github.com/yeyupiaoling/VoiceprintRecognition-PaddlePaddle)/[AudioClassification-PaddlePaddle](https://github.com/yeyupiaoling/AudioClassification-PaddlePaddle) 多年来的关注和建议，以及在诸多问题上的帮助。
 - 非常感谢 [mymagicpower](https://github.com/mymagicpower) 采用PaddleSpeech 对 ASR 的[短语音](https://github.com/mymagicpower/AIAS/tree/main/3_audio_sdks/asr_sdk)及[长语音](https://github.com/mymagicpower/AIAS/tree/main/3_audio_sdks/asr_long_audio_sdk)进行 Java 实现。
 - 非常感谢 [JiehangXie](https://github.com/JiehangXie)/[PaddleBoBo](https://github.com/JiehangXie/PaddleBoBo) 采用 PaddleSpeech 语音合成功能实现 Virtual Uploader(VUP)/Virtual YouTuber(VTuber) 虚拟主播。

--- a/paddlespeech/t2s/frontend/g2pw/bopomofo_to_pinyin_wo_tune_dict.json
+++ b/paddlespeech/t2s/frontend/g2pw/bopomofo_to_pinyin_wo_tune_dict.json
-{"ㄌㄧㄥ": "ling", "ㄩㄢ": "yuan", "ㄒㄧㄥ": "xing", "ㄑㄧㄡ": "qiu", "ㄊㄧㄢ": "tian", "ㄎㄨㄚ": "kua", "ㄨ": "wu", "ㄧㄣ": "yin", "ㄧ": "yi", "ㄒㄧㄝ": "xie", "ㄔㄡ": "chou", "ㄋㄨㄛ": "nuo", "ㄉㄢ": "dan", "ㄒㄩ": "xu", "ㄒㄩㄥ": "xiong", "ㄌㄧㄡ": "liu", "ㄌㄧㄣ": "lin", "ㄒㄧㄤ": "xiang", "ㄩㄥ": "yong", "ㄒㄧㄣ": "xin", "ㄓㄣ": "zhen", "ㄉㄞ": "dai", "ㄆㄢ": "pan", "ㄖㄨ": "ru", "ㄇㄚ": "ma", "ㄑㄧㄢ": "qian", "ㄘ": "ci", "ㄓㄨㄥ": "zhong", "ㄋㄟ": "nei", "ㄔㄥ": "cheng", "ㄈㄥ": "feng", "ㄓㄨㄛ": "zhuo", "ㄈㄤ": "fang", "ㄠ": "ao", "ㄗㄨㄛ": "zuo", "ㄓㄡ": "zhou", "ㄉㄨㄥ": "dong", "ㄙㄨ": "su", "ㄑㄩㄥ": "qiong", "ㄎㄨㄤ": "kuang", "ㄨㄤ": "wang", "ㄌㄟ": "lei", "ㄋㄠ": "nao", "ㄓㄨ": "zhu", "ㄕㄨ": "shu", "ㄕㄣ": "shen", "ㄐㄧㄝ": "jie", "ㄉㄧㄝ": "die", "ㄔ": "chi", "ㄌㄨㄥ": "long", "ㄧㄥ": "ying", "ㄅㄥ": "beng", "ㄌㄢ": "lan", "ㄇㄧㄠ": "miao", "ㄌㄧ": "li", "ㄐㄧ": "ji", "ㄩ": "yu", "ㄌㄨㄛ": "luo", "ㄔㄞ": "chai", "ㄏㄨㄣ": "hun", "ㄏㄨㄟ": "hui", "ㄖㄠ": "rao", "ㄏㄢ": "han", "ㄒㄧ": "xi", "ㄊㄞ": "tai", "ㄧㄠ": "yao", "ㄐㄩㄣ": "jun", "ㄌㄩㄝ": "lve", "ㄊㄤ": "tang", "ㄓㄠ": "zhao", "ㄓㄞ": "zhai", "ㄓㄚ": "zha", "ㄦ": "er", "ㄖㄢ": "ran", "ㄑㄧ": "qi", "ㄙㄜ": "se", "ㄙ": "si", "ㄙㄚ": "sa", "ㄎㄨㄟ": "kui", "ㄆㄨ": "pu", "ㄊㄚ": "ta", "ㄉㄨ": "du", "ㄊㄨ": "tu", "ㄧㄤ": "yang", "ㄡ": "ou", "ㄇㄧㄢ": "mian", "ㄨㄣ": "wen", "ㄉㄧㄠ": "diao", "ㄇㄧㄝ": "mie", "ㄨㄚ": "wa", "ㄋㄧㄠ": "niao", "ㄧㄡ": "you", "ㄔㄜ": "che", "ㄑㄩㄢ": "quan", "ㄘㄞ": "cai", "ㄌㄧㄤ": "liang", "ㄍㄨ": "gu", "ㄇㄠ": "mao", "ㄍㄨㄚ": "gua", "ㄙㄨㄟ": "sui", "ㄇㄢ": "man", "ㄕ": "shi", "ㄎㄡ": "kou", "ㄊㄧㄥ": "ting", "ㄅㄧㄥ": "bing", "ㄏㄨㄛ": "huo", "ㄍㄨㄥ": "gong", "ㄑㄧㄣ": "qin", "ㄐㄩㄥ": "jiong", "ㄌㄨ": "lu", "ㄋㄢ": "nan", "ㄅㄧ": "bi", "ㄑㄧㄚ": "qia", "ㄆㄧ": "pi", "ㄉㄧㄢ": "dian", "ㄈㄨ": "fu", "ㄍㄜ": "ge", "ㄅㄞ": "bai", "ㄍㄢ": "gan", "ㄒㄩㄢ": "xuan", "ㄌㄤ": "lang", "ㄕㄜ": "she", "ㄏㄨㄚ": "hua", "ㄊㄡ": "tou", "ㄆㄧㄢ": "pian", "ㄉㄧ": "di", "ㄖㄨㄢ": "ruan", "ㄜ": "e", "ㄑㄧㄝ": "qie", "ㄉㄡ": "dou", "ㄖㄨㄟ": "rui", "ㄘㄨㄟ": "cui", "ㄐㄧㄢ": "jian", "ㄔㄨㄥ": "chong", "ㄉㄥ": "deng", "ㄐㄩㄝ": "jue", "ㄒㄩㄝ": "xue", "ㄒㄧㄠ": "xiao", "ㄗㄢ": "zan", "ㄓㄢ": "zhan", "ㄗㄡ": "zou", "ㄘㄡ": "cou", "ㄔㄨㄚ": "chua", "ㄈㄟ": "fei", "ㄅㄟ": "bei", "ㄔㄨ": "chu", "ㄅㄚ": "ba", "ㄎㄨㄞ": "kuai", "ㄒㄧㄚ": "xia", "ㄏㄜ": "he", "ㄅㄧㄝ": "bie", "ㄌㄩ": "lv", "ㄙㄨㄢ": "suan", "ㄏㄥ": "heng", "ㄍㄨㄟ": "gui", "ㄌㄡ": "lou", "ㄊㄧ": "ti", "ㄌㄜ": "le", "ㄙㄨㄣ": "sun", "ㄒㄧㄢ": "xian", "ㄑㄩㄝ": "que", "ㄓ": "zhi", "ㄐㄧㄚ": "jia", "ㄏㄨ": "hu", "ㄌㄚ": "la", "ㄎㄜ": "ke", "ㄞ": "ai", "ㄨㄟ": "wei", "ㄏㄨㄢ": "huan", "ㄕㄨㄚ": "shua", "ㄕㄨㄤ": "shuang", "ㄍㄞ": "gai", "ㄏㄞ": "hai", "ㄧㄢ": "yan", "ㄈㄢ": "fan", "ㄆㄤ": "pang", "ㄙㄨㄥ": "song", "ㄋㄜ": "ne", "ㄔㄣ": "chen", "ㄍㄨㄛ": "guo", "ㄣ": "en", "ㄋㄍ": "ng", "ㄆㄚ": "pa", "ㄈㄚ": "fa", "ㄆㄡ": "pou", "ㄏㄡ": "hou", "ㄑㄩ": "qu", "ㄒㄩㄣ": "xun", "ㄋㄧㄝ": "nie", "ㄏㄨㄥ": "hong", "ㄊㄨㄣ": "tun", "ㄨㄞ": "wai", "ㄕㄡ": "shou", "ㄧㄝ": "ye", "ㄐㄩ": "ju", "ㄙㄡ": "sou", "ㄌㄨㄣ": "lun", "ㄋㄧㄚ": "nia", "ㄆㄣ": "pen", "ㄈㄣ": "fen", "ㄔㄨㄣ": "chun", "ㄋㄧㄡ": "niu", "ㄖㄡ": "rou", "ㄉㄨㄛ": "duo", "ㄗㄜ": "ze", "ㄕㄥ": "sheng", "ㄎㄨ": "ku", "ㄧㄚ": "ya", "ㄓㄨㄟ": "zhui", "ㄍㄡ": "gou", "ㄅㄛ": "bo", "ㄋㄚ": "na", "ㄒㄧㄡ": "xiu", "ㄘㄨ": "cu", "ㄎㄨㄛ": "kuo", "ㄌㄠ": "lao", "ㄘㄨㄥ": "cong", "ㄉㄚ": "da", "ㄆㄛ": "po", "ㄙㄞ": "sai", "ㄌㄥ": "leng", "ㄖㄨㄥ": "rong", "ㄋㄧ": "ni", "ㄆㄠ": "pao", "ㄎㄢ": "kan", "ㄨㄥ": "weng", "ㄨㄢ": "wan", "ㄏㄠ": "hao", "ㄐㄧㄥ": "jing", "ㄊㄢ": "tan", "ㄅㄨ": "bu", "ㄗㄤ": "zang", "ㄐㄧㄡ": "jiu", "ㄇㄟ": "mei", "ㄇㄨ": "mu", "ㄉㄨㄟ": "dui", "ㄅㄤ": "bang", "ㄅㄠ": "bao", "ㄔㄤ": "chang", "ㄓㄤ": "zhang", "ㄗㄨㄥ": "zong", "ㄍㄨㄣ": "gun", "ㄌㄧㄠ": "liao", "ㄔㄢ": "chan", "ㄓㄜ": "zhe", "ㄇㄥ": "meng", "ㄑㄧㄠ": "qiao", "ㄋㄤ": "nang", "ㄩㄣ": "yun", "ㄎㄞ": "kai", "ㄍㄠ": "gao", "ㄊㄠ": "tao", "ㄕㄢ": "shan", "ㄌㄞ": "lai", "ㄅㄢ": "ban", "ㄎㄨㄥ": "kong", "ㄔㄨㄛ": "chuo", "ㄋㄨ": "nu", "ㄆㄟ": "pei", "ㄆㄥ": "peng", "ㄘㄢ": "can", "ㄙㄨㄛ": "suo", "ㄊㄨㄥ": "tong", "ㄑㄧㄤ": "qiang", "ㄙㄠ": "sao", "ㄓㄨㄢ": "zhuan", "ㄢ": "an", "ㄔㄚ": "cha", "ㄕㄚ": "sha", "ㄌㄧㄢ": "lian", "ㄇㄧ": "mi", "ㄋㄡ": "nou", "ㄘㄠ": "cao", "ㄙㄣ": "sen", "ㄋㄣ": "nen", "ㄋㄧㄢ": "nian", "ㄇㄞ": "mai", "ㄩㄝ": "yue", "ㄋㄞ": "nai", "ㄏㄨㄞ": "huai", "ㄗ": "zi", "ㄌㄨㄢ": "luan", "ㄉㄧㄥ": "ding", "ㄇㄤ": "mang", "ㄋㄧㄥ": "ning", "ㄇㄧㄥ": "ming", "ㄗㄨㄟ": "zui", "ㄎㄤ": "kang", "ㄉㄜ": "de", "ㄅㄧㄢ": "bian", "ㄐㄧㄣ": "jin", "ㄔㄨㄟ": "chui", "ㄊㄨㄟ": "tui", "ㄗㄚ": "za", "ㄘㄣ": "cen", "ㄇㄧㄣ": "min", "ㄏㄨㄤ": "huang", "ㄗㄨ": "zu", "ㄘㄨㄛ": "cuo", "ㄊㄨㄛ": "tuo", "ㄑㄩㄣ": "qun", "ㄅㄧㄣ": "bin", "ㄊㄧㄠ": "tiao", "ㄍㄤ": "gang", "ㄉㄨㄢ": "duan", "ㄅㄧㄠ": "biao", "ㄉㄠ": "dao", "ㄖㄨㄣ": "run", "ㄐㄧㄠ": "jiao", "ㄨㄛ": "wo", "ㄘㄨㄢ": "cuan", "ㄖㄣ": "ren", "ㄇㄣ": "men", "ㄓㄨㄣ": "zhun", "ㄎㄨㄣ": "kun", "ㄔㄨㄤ": "chuang", "ㄗㄠ": "zao", "ㄓㄥ": "zheng", "ㄆㄧㄣ": "pin", "ㄅㄣ": "ben", "ㄐㄧㄤ": "jiang", "ㄐㄩㄢ": "juan", "ㄘㄥ": "ceng", "ㄏㄤ": "hang", "ㄋㄧㄣ": "nin", "ㄌㄧㄝ": "lie", "ㄍㄨㄤ": "guang", "ㄙㄢ": "san", "ㄊㄜ": "te", "ㄕㄨㄣ": "shun", "ㄕㄨㄟ": "shui", "ㄔㄠ": "chao", "ㄘㄜ": "ce", "ㄍㄨㄞ": "guai", "ㄎㄥ": "keng", "ㄕㄞ": "shai", "ㄉㄣ": "den", "ㄊㄨㄢ": "tuan", "ㄆㄧㄠ": "piao", "ㄑㄧㄥ": "qing", "ㄍㄥ": "geng", "ㄔㄨㄞ": "chuai", "ㄕㄠ": "shao", "ㄍㄣ": "gen", "ㄋㄨㄢ": "nuan", "ㄖㄥ": "reng", "ㄇㄡ": "mou", "ㄆㄞ": "pai", "ㄤ": "ang", "ㄎㄚ": "ka", "ㄍㄨㄢ": "guan", "ㄕㄨㄛ": "shuo", "ㄏㄣ": "hen", "ㄔㄨㄢ": "chuan", "ㄎㄨㄢ": "kuan", "ㄏㄟ": "hei", "ㄇㄛ": "mo", "ㄗㄞ": "zai", "ㄋㄥ": "neng", "ㄕㄨㄞ": "shuai", "ㄖㄜ": "re", "ㄋㄩ": "nv", "ㄆㄧㄥ": "ping", "ㄘㄤ": "cang", "ㄋㄨㄥ": "nong", "ㄎㄠ": "kao", "ㄗㄨㄢ": "zuan", "ㄎㄣ": "ken", "ㄍㄚ": "ga", "ㄗㄣ": "zen", "ㄉㄤ": "dang", "ㄗㄥ": "zeng", "ㄉㄨㄣ": "dun", "ㄘㄚ": "ca", "ㄖㄤ": "rang", "ㄘㄨㄣ": "cun", "ㄖㄨㄛ": "ruo", "ㄊㄧㄝ": "tie", "ㄊㄥ": "teng", "ㄙㄥ": "seng", "ㄖ": "ri", "ㄗㄨㄣ": "zun", "ㄋㄧㄤ": "niang", "ㄋㄩㄝ": "nve", "ㄙㄤ": "sang", "ㄓㄨㄤ": "zhuang", "ㄕㄤ": "shang", "ㄆㄧㄝ": "pie", "ㄕㄨㄢ": "shuan", "ㄈㄡ": "fou", "ㄉㄧㄡ": "diu", "ㄇㄜ": "me", "ㄈㄛ": "fo", "ㄌㄧㄚ": "lia", "ㄎㄟ": "kei", "ㄏㄚ": "ha", "ㄚ": "a", "ㄌㄛ": "lo", "ㄧㄛ": "yo", "ㄛ": "o", "ㄏㄋㄍ": "hng", "ㄋ": "n", "ㄌㄣ": "len", "ㄉㄧㄚ": "dia", "ㄇㄧㄡ": "miu", "ㄉㄟ": "dei", "ㄏㄇ": "hm", "ㄋㄨㄣ": "nun", "ㄓㄨㄞ": "zhuai", "ㄊㄟ": "tei", "ㄗㄟ": "zei", "ㄓㄨㄚ": "zhua", "ㄖㄨㄚ": "rua", "ê": "ê", "ㄟ": "ei", "ㄍㄟ": "gei", "ㄈㄧㄠ": "fiao", "ㄕㄟ": "shei", "ㄓㄟ": "zhei", "ㄥ": "eng", "ㄘㄟ": "cei", "ㄉㄧㄣ": "din", "ㄅㄧㄤ": "biang", "ㄧㄞ": "yai"}
\ No newline at end of file
--- a/paddlespeech/t2s/frontend/g2pw/char_bopomofo_dict.json
+++ b/paddlespeech/t2s/frontend/g2pw/char_bopomofo_dict.json
--- a/paddlespeech/t2s/frontend/g2pw/dataset.py
+++ b/paddlespeech/t2s/frontend/g2pw/dataset.py
+"""
+Credits
+ This code is modified from https://github.com/GitYCC/g2pW
+"""
 import numpy as np
 from paddlespeech.t2s.frontend.g2pw.utils import tokenize_and_map

@@ -128,8 +132,3 @@ def get_char_phoneme_labels(polyphonic_chars):
            char2phonemes[char] = []
        char2phonemes[char].append(labels.index(f'{char} {phoneme}'))
    return labels, char2phonemes
-
-
-def prepare_pos(pos_path):
-     return open(pos_path).read().rstrip().split('\n')
-
--- a/paddlespeech/t2s/frontend/g2pw/onnx_api.py
+++ b/paddlespeech/t2s/frontend/g2pw/onnx_api.py
+"""
+Credits
+ This code is modified from https://github.com/GitYCC/g2pW
+"""
 import os
 import json
 import onnxruntime
@@ -7,7 +11,10 @@ from opencc import OpenCC

 from paddlenlp.transformers import BertTokenizer

-from paddlespeech.t2s.frontend.g2pw.dataset import prepare_data, prepare_onnx_input, get_phoneme_labels, get_char_phoneme_labels
+from paddlespeech.t2s.frontend.g2pw.dataset import prepare_data,\
+                                                   prepare_onnx_input,\
+                                                   get_phoneme_labels,\
+                                                   get_char_phoneme_labels
 from paddlespeech.t2s.frontend.g2pw.utils import load_config

 MODEL_URL = 'https://paddlespeech.bj.bcebos.com/Parakeet/released_models/g2p/G2PWModel.tar'
@@ -34,6 +41,7 @@ def predict(session, onnx_input, labels):


 def download_model(model_dir):
+    os.makedirs(model_dir, exist_ok=True)
    wget_shell = "cd %s  && wget %s"%(model_dir,MODEL_URL)
    os.system(wget_shell)
    shell = "cd %s ;tar -xvf %s;cd %s/G2PWModel;rm -rf .*" % (model_dir,MODEL_URL.split("/")[-1], model_dir)
@@ -44,10 +52,12 @@ def download_model(model_dir):

 class G2PWOnnxConverter:
    def __init__(self, style='bopomofo', model_source=None, enable_non_tradional_chinese=False):
-        model_dir = os.path.dirname(os.path.abspath(__file__))
+        model_dir = os.path.join(os.path.expandvars('$HOME'), 'paddlespeech/models')
        if not os.path.exists(os.path.join(model_dir, 'G2PWModel/g2pW.onnx')):
            download_model(model_dir)

+        sess_options = onnxruntime.SessionOptions()
+        sess_options.intra_op_num_threads = 2
        self.session_g2pW =  onnxruntime.InferenceSession(os.path.join(model_dir, 'G2PWModel/g2pW.onnx'))
        self.config = load_config(os.path.join(model_dir, 'G2PWModel/config.py'), use_default=True)

@@ -65,16 +75,14 @@ class G2PWOnnxConverter:
        self.chars = sorted(list(self.char2phonemes.keys()))
        self.pos_tags = ['UNK', 'A', 'C', 'D', 'I', 'N', 'P', 'T', 'V', 'DE', 'SHI']

-        with open(os.path.join(os.path.dirname(os.path.abspath(__file__)),
-                               'bopomofo_to_pinyin_wo_tune_dict.json'), 'r',encoding='utf-8') as fr:
+        with open(os.path.join(model_dir,'G2PWModel/bopomofo_to_pinyin_wo_tune_dict.json'), 'r',encoding='utf-8') as fr:
            self.bopomofo_convert_dict = json.load(fr)
        self.style_convert_func = {
            'bopomofo': lambda x: x,
            'pinyin': self._convert_bopomofo_to_pinyin,
        }[style]

-        with open(os.path.join(os.path.dirname(os.path.abspath(__file__)),
-                               'char_bopomofo_dict.json'), 'r',encoding='utf-8') as fr:
+        with open(os.path.join(model_dir,'G2PWModel/char_bopomofo_dict.json'), 'r',encoding='utf-8') as fr:
            self.char_bopomofo_dict = json.load(fr)

        if self.enable_opencc:

--- a/paddlespeech/t2s/frontend/g2pw/utils.py
+++ b/paddlespeech/t2s/frontend/g2pw/utils.py
+
+"""
+Credits
+ This code is modified from https://github.com/GitYCC/g2pW
+"""
 import re
-import logging
 import sys

-
-class RunningAverage:
-    def __init__(self):
-        self.values = []
-
-    def add(self, val):
-        self.values.append(val)
-
-    def add_all(self, vals):
-        self.values += vals
-
-    def get(self):
-        if len(self.values) == 0:
-            return None
-        return sum(self.values) / len(self.values)
-
-    def flush(self):
-        self.values = []
-
-
 def wordize_and_map(text):
    words = []
    index_map_from_text_to_word = []
@@ -146,16 +130,4 @@ def load_config(config_path, use_default=False):
                for dict_k, dict_v in val.items():
                    if dict_k not in d:
                        d[dict_k] = dict_v
-    return config
-
-
-def get_logger(file_path):
-    logger = logging.getLogger()
-    logger.setLevel(logging.DEBUG)
-
-    output_file_handler = logging.FileHandler(file_path)
-    stdout_handler = logging.StreamHandler(sys.stdout)
-
-    logger.addHandler(output_file_handler)
-    logger.addHandler(stdout_handler)
-    return logger
+    return config
\ No newline at end of file
--- a/paddlespeech/t2s/frontend/zh_frontend.py
+++ b/paddlespeech/t2s/frontend/zh_frontend.py
@@ -79,7 +79,7 @@ class Frontend():
        self.tone_modifier = ToneSandhi()
        self.text_normalizer = TextNormalizer()
        self.punc = "：，；。？！“”‘’':,;.?!"
-        # g2p_model can be pypinyin and g2pM
+        # g2p_model can be pypinyin and g2pM and g2pW
        self.g2p_model = g2p_model
        if self.g2p_model == "g2pM":
            self.g2pM_model = G2pM()
@@ -87,6 +87,7 @@ class Frontend():
                with_tone=True, with_erhua=False)
        elif self.g2p_model == "g2pW":
            self.corrector = Polyphonic()
+            self.g2pM_model = G2pM()
            self.g2pW_model = G2PWOnnxConverter(style='pinyin', enable_non_tradional_chinese=True)
            self.pinyin2phone = generate_lexicon(
                with_tone=True, with_erhua=False)
@@ -180,8 +181,14 @@ class Frontend():
            initials = []
            finals = []
            seg_cut = self.tone_modifier.pre_merge_for_modify(seg_cut)
+            # 为了多音词获得更好的效果，这里采用整句预测
            if self.g2p_model == "g2pW":
-                pinyins = self.g2pW_model(seg)[0]
+                try:
+                    pinyins = self.g2pW_model(seg)[0]
+                except Exception:
+                    # g2pW采用模型采用繁体输入，如果有cover不了的简体词，采用g2pM预测
+                    print("[%s] not in g2pW dict,use g2pM"%seg)
+                    pinyins = self.g2pM_model(seg, tone=True, char_split=False)
                pre_word_length = 0
                for word, pos in seg_cut:
                    sub_initials = []