OpenAI unveiled its next-generation voice model, GPT Live, on Thursday, featuring a more natural-sounding ChatGPT voice.
GPT Live is engineered to make artificial intelligence (AI) conversations feel remarkably human-like. It listens to users’ voices in real-time, grasps context, and maintains a fluid conversational flow.
Previously, ChatGPT’s voice AI would convert speech to text, generate responses using a large language model (LLM), and then read the answers aloud. This process involved a slight delay.
In contrast, GPT Live continuously processes speech and understands conversational context on the fly. It can respond naturally while users are speaking and adapt seamlessly when users interrupt or pose follow-up questions.
Moreover, GPT Live separates the quick-reacting voice conversation feature from the complex reasoning capability. This design allows ChatGPT’s voice function to maintain natural dialogue while providing in-depth answers and handling requests that require tool usage.
The updated ChatGPT voice also enhances real-time translation. It can now continuously listen to and process users’ speech, offering immediate translations that align with the conversation’s flow. This significantly reduces the lag time between speech and translation.
OpenAI stressed that GPT Live marks a crucial advancement in both the naturalness and intelligence of voice AI. They aim to develop technology that seamlessly integrates tasks typically done via text or AI agent requests into voice conversations.
Looking ahead, GPT Live’s architecture is designed for expansion. It’s set to incorporate various backend models, new input and output methods, and application programming interface (API) based applications in the future.