OpenAI shared more about how it cast ChatGPT's voices

Voice Mode is one of ChatGPT's most beloved features, according to OpenAI. This may have contributed to the feature's recently announced revamp, now that GPT-4o has gone natively multimodal. In a recent blog post, OpenAI shared some insights into the five-month process that involved reviewing over 400 submissions to unveil the final five actors who would deliver the voices —Breeze, Cove, Ember, Juniper, and Sky— that have come to characterize ChatGPT's voice capabilities.

The process started in early 2023 when OpenAI partnered with independent, well-known, award-winning casting directors and producers to set a list of criteria that ensured each voice would have a unique personality and global appeal. The sought-after features included voices from multilingual actors or actors from diverse backgrounds, a feeling of timelessness, trust-inspiring voices that are natural and easy to listen to, and "warm, engaging, confidence-inspiring, charismatic voices with rich tone."

After publishing the call for talent, the casting team received over 400 submissions. Applicants auditioned by recording themselves reading a provided selection of ChatGPT responses, ranging from mindfulness and travel plan brainstorming to everyday conversations. After creating a list of 14 finalists, the casting team continued to refine their selections before presenting them to OpenAI. The finalists participated in a discussion with OpenAI representatives, where they were explained the technology's capabilities and limitations, the risks involved, and the safeguards in place so each actor understood the scope and intentions of the Voice Mode project before committing. Finally, after further consideration from a product perspective, an internal team selected the final five voices of Breeze, Cove, Ember, Juniper, and Sky.

OpenAI has remarked that none of the voices were selected because of their similarity with a celebrity's voice, and all are the talent's natural speaking voices. Moreover, actors are compensated above the top-of-market rate for as long as the voices are used in OpenAI's products. Breeze, Cove, Ember, Juniper, and Sky will likely be featured in the upcoming Voice Mode for GPT-4o alpha, which will roll out for ChatGPT Plus users in the coming weeks. In the new Voice Mode, GPT-4o will showcase its capabilities to handle interruptions, manage group conversations, filter background noise, and adapt to the set tone. Farther down the line, OpenAI plans to introduce additional voices that better reflect the diversity of its users.