Firstly, why are the banks making a big deal out of this fancy new way of authenticating ourselves when A, chip and PIN is protected by a lousy four digit PIN (come on, we're expected to know eleven digit phone numbers, can't they offer us the option to have a longer PIN if we want?) and B, we are expected to take it on trust that the little keypad connects to a proper secure device for the purposes of completing the transaction. I, personally, do not believe the so-called security one bit. There is no way to tell if the device isn't harvesting card numbers and PINs. In today's world of instant communications, it strikes me as odd that inserting your card can't open a secure connection to your bank, with the bank passing a special security phrase that can be shown on the LCD (an improbable passphrase of YOUR choosing known only to the bank such as "Duck! Duck! The turd is shiny giraffe!") so you know it is supposedly safe - this being an encrypted channel between the device itself and the bank. All the till will ever see is a yes/no response and some gibberish passing to and fro.
Secondly, I know Sneakers is an old film, but it is perhaps the best example in cinema of how to pwn voice based authentication systems. Do not forget that if this is done on a computer or smartphone it will likely require specific software (which is a whole other trust level - such as NatWest's constant demands for me to install Trusteer Rapport on my computers) and if it is done over the phone you'll be subject to the limitations of POTS, which may well mean that the authentication could be able to be fooled by a cassette tape recording of your voice.
So, sorry, call me cynical, but I'm wondering what this development is really intended to address. I'm wondering how it could be (ab)used to push more liability onto the end-user (who, helium balloons and singing aside, don't usually get the ability to change their voice...or other biometrics).