OpenAI is rolling out Advanced Voice Mode to a select group of ChatGPT Plus users. This feature enables more natural, real-time conversations, allows interruptions, and senses and responds to emotions. Users in this alpha phase will receive instructions via email and a message in their mobile app. The rollout will continue gradually, with all Plus users expected to have access by fall. Video and screen sharing capabilities will be introduced later.
Safety and Quality Measures
OpenAI has focused on reinforcing the safety and quality of voice conversations. They tested GPT-4o's voice capabilities with over 100 external red teamers across 45 languages. To protect privacy, the model is trained to speak in only four preset voices, and systems are in place to block outputs that deviate from these voices. Additional guardrails prevent requests for violent or copyrighted content. Insights from this alpha phase will help improve the safety and enjoyment of the Advanced Voice experience. A detailed report on GPT-4o’s capabilities, limitations, and safety evaluations will be shared in early August.
The initial rollout of ChatGPT’s Advanced Voice Mode began on Tuesday, providing users their first access to GPT-4o’s hyperrealistic audio responses. The alpha version is available to a small group of ChatGPT Plus users, with a full rollout to all Plus users expected by fall 2024.
Controversy and Adjustments
When OpenAI first showcased GPT-4o’s voice in May, it impressed audiences with its quick responses and human-like voice, particularly one resembling Scarlett Johansson. Johansson denied giving permission for her voice to be used and hired legal counsel. OpenAI denied using her voice but later removed the voice from its demo. The release of Advanced Voice Mode was delayed to improve safety measures.
Future Features
Video and screensharing capabilities, showcased during OpenAI’s Spring Update, will not be part of this alpha and will be launched at a later date. For now, some premium users will have access to the voice feature demonstrated earlier.
