The video provides the best introduction:

Basically, you can now have your typist create a transcript of your video and then directly upload that to YouTube, which will use speech recognition to time-align the transcript and turn it into captions. That’s awesome! You can even download those captions and keep them for some other purpose, such as add them to your local archive of videos.

For some select partners, YouTube is even experimenting with a fully automated captioning system, where the transcript is created using speech recognition. Since we know the quality of Google voice from searches performed through Google searches on mobile phones, let’s be skeptical of the quality of the fully automated solution. But it is certainly a great start and we can certainly expect Google to continue making improvements to their speech recognition technology.