https://arxiv.org/abs/2111.00161
Pseudo-Labeling for Massively Multilingual Speech Recognition (Loren Lugosch, Tatiana Likhomanenko, Gabriel Synnaeve, Ronan Collobert)
asr 모델에 60개 언어 집어넣기. multilingual pretraining -> monlingual finetuning -> pseudo labeling 순서로군요.
#asr #semi_supervised_learning #multilingual