OpenAI releases voice and photo functionality for ChatGPT Plus and Enterprise

By admin On Sep 25, 2023

Spread the love

OpenAI announces two new features for ChatGPT: the ability to process images and the ability to recognize voices and ‘talk’ back. The functionality will be rolled out over a two-week period for Plus and Enterprise subscribers.

The new speech feature allows users to talk to ChatGPT, after which OpenAI’s language model talks back using a text-to-speech model. In principle, all applications of the chatbot are also possible via the voice function; users can request stories, explanations, recipes, tips and concrete information, and now receive this in audio form. The audio function is opt-in and will only be available for the iOS and Android versions of the chatbot. This requires a Plus or Enterprise subscription.

The company recommends users to speak to ChatGPT in English. Some other languages could also be processed by the language model, but the results would still be best in English. OpenAI has based the five available voices on real voice actors, but can in principle imitate any voice based on a few sentences. Earlier today, Spotify announced that the music service is using the same technology to translate podcasts.

Furthermore, the company is announcing a photo feature that will allow users to upload images, documents and photos, and ChatGPT can then respond dynamically based on this visual information. Some examples are asking for technical help when adjusting a bicycle, requesting recipes based on the contents of the refrigerator or analyzing a graph. This functionality will be available for all platforms in the next two weeks, again provided that a Plus or Enterprise subscription is purchased.

OpenAI states that both functions also carry risks. For example, malicious parties can theoretically use the speech function to imitate real people. Therefore, the company limits the technology to use within ChatGPT’s voice chat feature. Furthermore, the developer says that the image feature is “significantly limited” when it comes to responding to real people.