| The abstract from the paper is the following: | |
| We show for the first time that learning powerful representations from speech audio alone followed by fine-tuning on | |
| transcribed speech can outperform the best semi-supervised methods while being conceptually simpler. |