Fetching the paper…

Improving non-autoregressive end-to-end speech recognition with pre-trained acoustic and language models · Around