if it's the former case the link won't help you, but I'd suggest you reconsider your approach. Why would you use voice AI technology to capture DTMF inputs from a user?