May, 2024
跨模态离散语音单元的 ASR 错误修正
Crossmodal ASR Error Correction with Discrete Speech Units
Yuanchao Li, Pinzhen Chen, Peter Bell, Catherine Lai
TL;DRASR Error Correction (AEC) is investigated to tackle the Low-Resource Out-of-Domain (LROOD) problem, covering pre-training, fine-tuning, and incorporating discrete speech units to improve AEC quality. The proposed AEC approach demonstrates feasibility, efficacy, generalizability, superiority, and suitability for downstream applications.