OpenAI announced that it has developed an artificial intelligence model that can clone a voice with a 15-second speech recording. The company will discuss the technology, Voice Engine, with authorities and experts before making it public.

OpenAI introduced Voice Engine, an artificial intelligence model that could revolutionise voice cloning. This model, the product of two years of development, can clone any voice just by listening to a 15-second audio recording. This is indeed a frightening progress.

According to the company, Voice Engine was trained with licensed audio recordings and public databases. Therefore, there are no copyright issues. OpenAI has absolutely no plans to make this technology widely available at this time because of the dangers of misuse.

In the blog post written by OpenAI, there is information about what this technology was developed for. These include providing reading assistance to people who are visually impaired or have reading difficulties, translating and dubbing for people who speak different languages, helping those with speech difficulties, providing new tools to content producers, and creating new research opportunities in areas such as language acquisition and speech therapy.
 

How does Voice Engine work?

After listening to a recording for 15 seconds, Voice Engine analyses the pitch and other voice characteristics to produce a synthetic voice that is very similar to the original voice. The fact that the artificial intelligence model has not been released for general use already shows that the synthetic voice is very close to the original.

The potential dangers of Voice Engine are quite frightening. There are risks such as fraud, identity theft, disinformation or creating deepfake videos through voice imitation. OpenAI says it is working to address these concerns and plans to consult with "authorities and experts" before making Voice Engine publicly available.

The development of Voice Engine also raises important ethical and legal questions. How to control such powerful artificial intelligence models and protect them from abuse will be an important issue to be addressed in the coming years. You can listen to examples of cloned voices on OpenAI's blog page.