The magic about the @OpenAI realtime API is that you can go from voice to voice without a step through in text. Now, at many companies, hundreds of call center roles can be replaced by a few engineering roles. ---- 👉 Previously [Audio Input] -> [Automatic Speech Recognition (e.g., Whisper)] | v [Transcribed Text] -> [Text Model for Inference/Reasoning] | v [Text Output] -> [Text-to-Speech Model] | v [Audio Output] 👉 Now [Audio Input] -> [Chat Completions API] -> [Audio Output] ----- Under the hood, the Realtime API lets you create a persistent WebSocket connection to exchange messages with GPT-4o. The API supports function calling(opens in a new window), which makes it possible for voice assistants to respond to user requests by triggering actions or pulling in new context.