OPENAI RELEASES GPT-LIVE-1 CONVERSATIONAL VOICE MODELS

OpenAI released two new conversational voice models on Wednesday, designated GPT-Live-1 and GPT-Live-1 mini, according to the company. The models employ full-duplex technology, enabling simultaneous speech and listening capabilities. This architecture permits users to interrupt naturally during conversations and facilitates real-time translation features. OpenAI stated that the new models produce more natural-sounding speech and handle turn-taking more effectively than previous iterations.

The company is replacing the existing Advanced Voice Mode in ChatGPT with GPT-Live-1 mini as the default option for all users. Subscribers to paid tiers will gain access to the larger GPT-Live-1 model. The earlier voice system operated through a three-stage pipeline: a speech-to-text component transcribed user speech, a large language model generated responses, and a text-to-speech component delivered the final output. The new full-duplex architecture consolidates these functions to allow simultaneous input and output.

OpenAI described GPT-Live-1 as delivering more natural conversations without interruptions. The rollout commenced today across ChatGPT's user base. The company did not specify a timeline for complete deployment or announce any other planned features relating to the new voice models.