Naver-AI-Hackathon / AI-Speech

31 stars 5 forks source link

nsml test data에 대한 CER 계산 방식 #53

Closed alberto-hong closed 4 years ago

alberto-hong commented 4 years ago

제가 이런 고민을 하는 이유는 model의 output level을 character level에서 다른 level로 바꿔보려고 하는데(그러면 dictionary가 완전히 바뀌게 되는데), 이게 배포해주신 main.py의 infer() 을 보면 test data에 대해서 string type의 sentence로 prediction을 하게 되는데,

test data의 ground truth transcription은 sentence로 바로 불러오는 건가요? 아니면, hackathon.labels을 이용해서 다시 바꾸게 되나요?