pip install -r requirements.txt

mkdir saved_models
cd saved_models
wget https://storage.googleapis.com/esun-ai/bert_detection.zip
unzip bert_detection.zip
cd ..

python src/test_phonetic_mlm.py --config configs/config_phonetic_mlm.py --json data/aishell3_test.json

python src/predict_phonetic_mlm.py --config configs/config_phonetic_mlm.py --text_path misc/demo.txt

python src/train_typo_detector.py --config configs/config_detect.py

python src/test_typo_detector.py --config configs/config_detect.py --checkpoint saved_models/bert_detection/best_f1.pth --json data/aishell3_test.json

python src/predict_typo_detector.py --config configs/config_detect.py --checkpoint saved_models/bert_detection/best_f1.pth --text_path misc/demo.txt

@inproceedings{chen-etal-2021-integrated,
    title = "Integrated Semantic and Phonetic Post-correction for {C}hinese Speech Recognition",
    author = "Chen, Yi-Chang and Cheng, Chun-Yen and Chen, Chien-An and Sung, Ming-Chieh and Yeh, Yi-Ren",
    booktitle = "Proceedings of the 33rd Conference on Computational Linguistics and Speech Processing (ROCLING 2021)",
    month = oct,
    year = "2021",
    address = "Taoyuan, Taiwan",
    publisher = "The Association for Computational Linguistics and Chinese Language Processing (ACLCLP)",
    url = "https://aclanthology.org/2021.rocling-1.13",
    pages = "95--102",
    abstract = "Due to the recent advances of natural language processing, several works have applied the pre-trained masked language model (MLM) of BERT to the post-correction of speech recognition. However, existing pre-trained models only consider the semantic correction while the phonetic features of words is neglected. The semantic-only post-correction will consequently decrease the performance since homophonic errors are fairly common in Chinese ASR. In this paper, we proposed a novel approach to collectively exploit the contextualized representation and the phonetic information between the error and its replacing candidates to alleviate the error rate of Chinese ASR. Our experiment results on real world speech recognition datasets showed that our proposed method has evidently lower CER than the baseline model, which utilized a pre-trained BERT MLM as the corrector.",
}