An Effective Learning Method for Automatic Speech Recognition in Korean CI Patients’ Speech

The automatic speech recognition (ASR) model usually requires a large amount of training data to provide better results compared with the ASR models trained small data. It is difficult apply non-standard such as that cochlear implant (CI) patients, owing privacy concerns or difficulty access. In this paper, an effective finetuning and augmentation proposed. Experiments compare character error rate (CER) after basic proposed method. method achieved CER 36.03% on CI patient’s test dataset using only 2 h 30 min data, which 62% improvement over