The gpt realtime model sounds distorted af when calling on a free vapi number, is there a fix planned for this?