Tag
This paper proposes a continual learning approach to integrate disfluency tokens into pretrained ASR models, addressing catastrophic forgetting and improving recognition of disfluent speech.