It's insanely easy for a program to recognize individual digits and then type them out. Saying a name would also be trivial (recorded, generated, etc). Audio captchas have existed for a while, they have been an alternative for a long time, and they are even more annoying than images.

What we need is an abuse prevention system involving authentication. You can get reported for robo calls, and if enough different accounts you called have reported you, you get reviewed/banned/throttled. Let's not reinvent the wheel, it's been the standard way to deal with spam all over the web. But it requires rethinking the way we use phones, and a whole new infrastructure. It also brings some anonymity concerns but half-measures are just an arms race.