OpenAI has released an upgraded version of its AI voice assistant, called "Voice Mode Advanced," on September 24. This comes just two months after the launch of "Voice Mode Standard" in late July, and it is available to existing ChatGPT subscribers. The AI voice assistant utilizes generative AI to carry out complex user commands, including real-time conversations and internet searches.The "Advanced" version adds five new voices, trained by professional voice actors, and has improved upon the awkward phrasing noted in Korean and Japanese from the previous version, bringing it closer to native pronunciation. I had the opportunity to test "Voice Mode Advanced" on September 23. As soon as the conversation began, I felt as though I was chatting with a "real Korean," thanks to the natural tone of voice. When I asked, "Could you introduce yourself in a cute manner?" the AI immediately elongated its phrases and raised its pitch, delivering a charming performance.

Joan Zhang, Lead of Model Behavior at OpenAI, explained that they refined the voice by gathering feedback from a diverse group, including Korean employees within the company. When asked about the upcoming U.S. presidential election in November, or thoughts on former President Trump, the AI cautiously replied, "It's a sensitive topic," emphasizing the variety of opinions on such matters. OpenAI suggested that the voice AI could serve as a conversational partner or an idea brainstormer. For example, when I inquired, "I have a day trip planned in Las Vegas; what should I see?" the AI recommended, "If you’re short on time, it’s best to focus on just three attractions," suggesting experiences like the fountain show at the Venetian. Jackie Shannon, Lead of Multimodal Safety at OpenAI, noted that they have implemented safety measures to prevent the AI from generating violent or copyright-infringing content and to ensure it cannot mimic the voices of specific individuals.

(The Chosun Daily, September 25, 2024)