A speaker announcing a Sendspin pairing code should not have to play audio chosen by a stranger. The protocol removes a mechanism that let a server supply recordings of the ten digits before the two devices had established trust. Speaking clients instead use their own bundled voice, selecting its language using the server's language hint.
Chris's revision reverses an August specification decision that removed the need to store those voice clips on the speaker. Approved by maximmaxim345, the revised rule means an unauthenticated peer no longer gets to choose audio for the speaker or send audio into its decoder through this pairing mechanism.