Lacking of 60k data

[SkyYap]

I have read through the paper and found out they are using 60k unlabeled audio data to train the AI.

Anyone has any direction we can reproduce the method in the paper that using speech recognition to generate the transcription and increase the accuracy of it?

[eonglints]

Hey, just to say that the Multilingual LibriSpeech dataset would do the trick. It's also from LibriVox and has 45k hours of transcribed speech.

[SkyYap]

Hey, just to say that the Multilingual LibriSpeech dataset would do the trick. It's also from LibriVox and has 45k hours of transcribed speech.

thanks for your sharing! Appreciate it.