OpenAI has started testing its "Advanced Voice Mode" with a select group of ChatGPT Plus subscribers.
This new feature aims to enable more natural real-time conversations, allowing users to interrupt the AI at any point. The company is gradually rolling out the feature, notifying chosen users via email and in-app messages with instructions. OpenAI plans to expand access to all ChatGPT Plus users by fall, with video and screen sharing capabilities to follow later.
Video: OpenAI
OpenAI tested the voice capabilities of GPT-4o, the model powering Advanced Voice Mode, with over 100 external red teams across 45 languages. To protect user privacy, the model speaks only in four preset voices. The company has also implemented systems to prevent output that deviates from these voices.
The initial announcement of Voice Mode in May sparked controversy over the computer-generated voice, which reportedly was supposed to sound like actress Scarlett Johansson without proper licensing.
The alpha test, originally scheduled for late June, was delayed by a month for security reasons. Insights from this phase will help refine and secure the Advanced Voice experience.
