This is the code and instruction for the paper Interventional Speech Noise Injection for ASR Generalizable Spoken Language Understanding.
Our code is based on the code of the paper ASR error correction with constrained decoding on operation prediction. 
1. clone the code from the github https://github.com/yangjingyuan/ConstDecoder
2. place the codes in this directory identical to the cloned code.
3. make phoneme label with phoneme_softlabel_clean.ipynb
4. train SNI model with train_constrained_noisygen_phoneme.py
5. run generation with phoneme_random_phoneme_generate.py