GitHub - GitYCC/g2pW: Chinese Mandarin Grapheme-to-Phoneme Converter. 中文轉注音或拼音 (INTERSPEECH 2022) (original) (raw)

g2pW: Mandarin Grapheme-to-Phoneme Converter

Authors: Yi-Chang Chen, Yu-Chuan Chang, Yen-Cheng Chang and Yi-Ren Yeh

This is the official repository of our paper g2pW: A Conditional Weighted Softmax BERT for Polyphone Disambiguation in Mandarin (INTERSPEECH 2022).

News

• g2pW is included in PaddlePaddle/PaddleSpeech
• g2pW is included in mozillazg/pypinyin-g2pW

Getting Started

Dependency / Install

(This work was tested with PyTorch 1.7.0, CUDA 10.1, python 3.6 and Ubuntu 16.04.)

• Install PyTorch
• $ pip install g2pw

Quick Demo

from g2pw import G2PWConverter conv = G2PWConverter() sentence = '上校請技術人員校正FN儀器' conv(sentence) [['ㄕㄤ4', 'ㄒㄧㄠ4', 'ㄑㄧㄥ3', 'ㄐㄧ4', 'ㄕㄨ4', 'ㄖㄣ2', 'ㄩㄢ2', 'ㄐㄧㄠ4', 'ㄓㄥ4', None, None, 'ㄧ2', 'ㄑㄧ4']] sentences = ['銀行', '行動'] conv(sentences) [['ㄧㄣ2', 'ㄏㄤ2'], ['ㄒㄧㄥ2', 'ㄉㄨㄥ4']]

Load Offline Model

conv = G2PWConverter(model_dir='./G2PWModel-v2-onnx/', model_source='./path-to/bert-base-chinese/')

Support Simplified Chinese and Pinyin

from g2pw import G2PWConverter conv = G2PWConverter(style='pinyin', enable_non_tradional_chinese=True) conv('然而，他红了20年以后，他竟退出了大家的视线。') [['ran2', 'er2', None, 'ta1', 'hong2', 'le5', None, None, 'nian2', 'yi3', 'hou4', None, 'ta1', 'jing4', 'tui4', 'chu1', 'le5', 'da4', 'jia1', 'de5', 'shi4', 'xian4', None]]

Scripts

$ git clone https://github.com/GitYCC/g2pW.git

Train Model

For example, we train models on CPP dataset as follows:

$ bash cpp_dataset/download.sh
$ python scripts/train_g2p_bert.py --config configs/config_cpp.py

Testing

$ python scripts/test_g2p_bert.py \
    --config saved_models/CPP_BERT_M_DescWS-Sec-cLin-B_POSw01/config.py \
    --checkpoint saved_models/CPP_BERT_M_DescWS-Sec-cLin-B_POSw01/best_accuracy.pth \
    --sent_path cpp_dataset/test.sent \
    --output_path output_pred.txt

Prediction

$ python scripts/predict_g2p_bert.py \
    --config saved_models/CPP_BERT_M_DescWS-Sec-cLin-B_POSw01/config.py \
    --checkpoint saved_models/CPP_BERT_M_DescWS-Sec-cLin-B_POSw01/best_accuracy.pth \
    --sent_path cpp_dataset/test.sent \
    --lb_path cpp_dataset/test.lb

Checkpoints

• G2PWModel-v2.zip
• G2PWModel-v2-onnx.zip

Citation

To cite the code/data/paper, please use this BibTex

@inproceedings{chen22d_interspeech, title = {g2pW: A Conditional Weighted Softmax BERT for Polyphone Disambiguation in Mandarin}, author = {Yi-Chang Chen and Yu-Chuan Steven and Yen-Cheng Chang and Yi-Ren Yeh}, year = {2022}, booktitle = {Interspeech 2022}, pages = {1926--1930}, doi = {10.21437/Interspeech.2022-216}, issn = {2958-1796}, }

Star History