In the past year, Mozilla Research announced an intiative to help machines learn how people speak - The Common Voice Project.
What you do is record a sentence they provide to your microphone and it'll get saved. Then, Mozilla will ask others on the Internet to validate the sentence, a simple yes or no if the sentence was read correctly. They're crowdsourcing the largest open collection of voice data!
I'm a big fan of this initiative. Having an open dataset of recorded voices matched to sentences opens up the doors for others on the web to try their hands at voice input. It gives everyone the ability to train voice recognition models which they could use to power their websites.
At the time of writing, Mozilla Research has validated 253 hours so far. Their goal is to hit 500 hours of validated hours next.
If youre interested, head on over to the Common Voice Project home page and to contributing your voice to the open web!