| By pretraining on large amounts of speech data, finetuning the model on only one hour of labeled speech data in a low-resource language can still produce high-quality results compared to previous ASR systems trained on 100x more labeled data. |
| By pretraining on large amounts of speech data, finetuning the model on only one hour of labeled speech data in a low-resource language can still produce high-quality results compared to previous ASR systems trained on 100x more labeled data. |