I just ran a test call where the model output the correct tool call and content (dtmf: 12345) within 1s of the request ("What is your zip code?"), but the dtmf tones did not come through over the phone for nearly 14 seconds. Can you give context/visibility to the flow between model output (function call) taking place and the dtmf being audibly submitted to the other line so we can potentially figure out how to reduce that window? @Sahil