OpenAI is gradually rolling out a new advanced voice mode for ChatGPT to a limited number of ChatGPT Plus subscribers. This feature, first showcased at the GPT-4 launch event in May, was criticized for sounding too much like Scarlett Johansson and was delayed for safety reasons.
The new voice mode demonstrated notable improvements over ChatGPT’s current voice capabilities at the event. OpenAI employees could interrupt and request different storytelling styles from the chatbot, which adeptly adjusted its responses in real-time.
Initially slated for an alpha release in late June, OpenAI postponed the rollout by a month to ensure the feature met its quality standards. The delay enhanced the model’s ability to detect and refuse inappropriate content.
According to OpenAI spokesperson Taya Christianson, the voice model was tested by over 100 external red teamers and experts trying to find technological weaknesses. Given the recent scrutiny of OpenAI’s safety policies, this pause was deemed necessary. Christianson also mentioned the addition of new filters designed to recognize and block requests to generate music or other copyrighted audio.
One of the main criticisms during the event was the resemblance of the onstage voice, “Sky,” to Scarlett Johansson’s character in the movie *Her*. Although the voice had been part of ChatGPT before OpenAI’s spring demo, it was removed after Johansson inquired about its creation.
Christianson assured that the new mode would use only four preset voices created with voice actors, emphasizing that ChatGPT could not impersonate other people’s voices, whether individuals or public figures, and would block any outputs that deviate from these preset voices.
OpenAI plans to extend the new voice mode to all ChatGPT Plus users in the fall, aiming to offer an enhanced and safer voice interaction experience.
