OpenAI is discontinuing the use of a voice, named Sky, in its ChatGPT due to its resemblance to actress Scarlett Johansson. The decision was made after users pointed out the similarity during a demonstration of the new GPT-4o model. OpenAI clarified that Sky's voice is not an imitation of Johansson's but belongs to another professional actress.
The voice gained attention on social media, with some users criticizing it for being overly flirtatious. Comparisons were made to Johansson's role in the 2013 film "Her", where she voiced a seductive virtual assistant. OpenAI CEO, Sam Altman, also referenced "Her" in a tweet after the event.
The demonstration was meant to highlight ChatGPT's improved conversational skills. However, it gained viral attention due to Sky's voice, which often responded with giggles and flirtatious comments.
ChatGPT's Voice Mode offers five distinct voices: Breeze, Cove, Ember, Juniper, and Sky. These voices were selected through a rigorous five-month process involving professionals from the voice acting industry, including talent agencies, casting directors, and industry advisors.
In September 2023, voice capabilities were introduced to ChatGPT, with the voices being sampled from professional voice actors. The actors were compensated above top-of-market rates, and this will continue as long as their voices are used in OpenAI's products. The company maintains that AI voices should not mimic celebrities, and to respect privacy, the names of the voice talents are not disclosed.
OpenAI partnered with award-winning casting directors and producers to create criteria for the voices. They looked for actors from diverse backgrounds, voices that felt timeless, approachable, warm, engaging, and easy to listen to.
In May 2023, a call for talent was issued, resulting in over 400 submissions from voice and screen actors. The casting team reviewed these submissions and selected an initial list of 14 actors. After discussions about the vision for human-AI voice interactions, the technology's capabilities, limitations, and risks, the final five voices were selected.
The actors were flown to San Francisco for recording sessions, and their voices were launched into ChatGPT in September 2023.
