OpenAI has started implementing the Advanced Voice Mode of ChatGPT, with hyper-realistic audio responses thanks to GPT-4o. According to TechCrunch, the alpha version is available to a small group of Plus users since Tuesday, and will be released to all users in the fall of 2024.
Initially launched in May, the feature surprised the whole world by having a voice called Sky that was very similar to Scarlett Johansson’s in the movie Her, although the actress denied authorizing the use of her voice and took legal action. OpenAI denied using her voice, but ended up removing Sky from the voice library.
This new feature will allow ChatGPT to speak and listen more smoothly, thanks to the multimodal capability of GPT-4o. Unlike the previous voice mode, which used three separate models, the new system is able to process the entire conversation without the help of auxiliary models, significantly reducing latency and allowing for the perception of emotional intonations.
The launch is being done gradually to monitor the usage of the new feature, and selected users will receive notifications through the app and via email with instructions to test it. Since the May demo, OpenAI has tested the voice capabilities with over 100 security experts who speak 45 languages. The company will publish a report in early August about these tests.
The available voices are Juniper, Breeze, Cove, and Ember, created in collaboration with voice actors, while Sky is no longer available. Lindsay McCallum, spokesperson for OpenAI, assures that ‘ChatGPT cannot replace the voices of other people’ and states that filters will be implemented to avoid generating music or other copyrighted content. With this measure, the company seeks to avoid future legal problems and controversies like those that have occurred with other voice cloning technologies.
