Where and in what form is the voice recording sent when using speech recognition via the Web Speech API?

Question

Google Chrome

When recognizing speech using the Web Speech API, the voice recording is sent somewhere to the Google servers and from there the text into which the speech has been converted already comes. I read that speech is sent via WebSockets. How, in what format, and where (to which URL) is the voice recording sent?
A sniffer is not an option to look, everything is encrypted there.

Answer 1

Here's the latest on Google's Web Speech API.