OpenAI ChatGPT Voice Selection Process

OpenAI has detailed the professional casting process used to select the voices for ChatGPT, emphasizing that the voices are sampled from professional actors who are compensated above top-of-market rates. This transparency follows a public dispute regarding the voice "Sky" and its resemblance to actress Scarlett Johansson.

Casting Criteria and Selection Process

OpenAI partnered with independent, award-winning casting directors and producers in early 2023 to establish criteria for ChatGPT's voices. The goal was to identify voices that were approachable, timeless, and inspired trust, with a specific focus on characteristics such as:

  • Diversity: Actors from diverse backgrounds or those capable of speaking multiple languages.
  • Tone: Warm, engaging, charismatic, and confidence-inspiring with a rich tone.
  • Accessibility: Natural and easy to listen to.

Following a call for talent issued on May 10, 2023, the casting agency received over 400 submissions. Actors auditioned by recording scripts featuring ChatGPT responses, ranging from mindfulness exercises to travel planning.

Timeline of Voice Implementation

The selection and recording process spanned five months, leading to the launch of the initial voice set in September 2023. The timeline of events is as follows:

  • May 10, 2023: Call for talent issued.
  • May 2023: Casting team reviewed and hand-selected 14 actors, which were further refined before being presented to OpenAI.
  • June and July 2023: Selected actors traveled to San Francisco for recording sessions and meetings with product and research teams.
  • September 11, 2023: Sam Altman contacted Scarlett Johansson to discuss her potential involvement as a sixth voice actor. She declined the opportunity one week later.
  • September 25, 2023: The voices for Breeze, Cove, Ember, Juniper, and Sky were launched in ChatGPT.
  • May 10, 2024: Sam Altman contacted Ms. Johansson's team again to ask if she would reconsider joining as a future voice for GPT-4o.
  • May 13, 2024: GPT-4o was introduced.
  • May 19, 2024: OpenAI paused the use of the "Sky" voice out of respect for Ms. Johansson's concerns.

Voice Actor Compensation and Privacy

OpenAI states that each voice actor receives compensation above top-of-market rates, and this compensation will continue for as long as their voices are used in the products. To protect the privacy of the voice talents, OpenAI does not share their names publicly.

Integration with GPT-4o

The original voice actors continue to collaborate with OpenAI for audio research and new capabilities in GPT-4o. The new Voice Mode for GPT-4o, currently entering alpha for ChatGPT Plus users, introduces improvements in natural interaction, including the ability to handle interruptions, manage group conversations, filter background noise, and adapt to tone.

Sources