Hi Pat, This is a known issue with OpenAI Realtime models over phone calls — realtime voices can sound slow/deep or robotic on PSTN calls while sounding fine in the web app. Short guidance: • Cause: it's an artifact of the OpenAI Realtime pipeline when routed to phone networks (codec/sampling/packetization + realtime voice rendering). See our Realtime guide: https://docs.vapi.ai/openai-realtime • Short-term workaround: switch the assistant to a non-realtime model (e.g., gpt-4o or gpt-realtime → gpt-4o) and use a non-realtime TTS voice (Vapi, ElevenLabs, Minimax) for cleaner audio over phone. • If you must keep realtime timing: use realtime for inference but render final audio with a non-realtime TTS and stream that to the phone (hybrid approach). • Next steps I can take: check your call ID and model/voice config in the dashboard Observe > Call Logs and confirm if you used a realtime model + OpenAI voice; if you share the call ID I’ll pull the details. Which call ID should we check?