OpenAI unveils revolutionary tool for cloning voices

After the presentation of Sora, an artificial intelligence (AI) that can generate a video from text, OpenAI unveils its new tool for cloning voice. This is Voice Engine which can reproduce a person’s voice, regardless of their accent.

OpenAI, the artificial intelligence giant, strikes again. After text, image and video, he is interested in voice. On March 29, 2024, the Californian company presented its new tool called Voice Engine, which has the ability to reproduce the human voice. That said, this innovation arouses as much wonder as concern.

Voice Engine only needs a 15 second sample!

After ChatGPT and Dall-E, OpenAI lifts the veil on its recent AI model, in this case, Voice Engine. A tool whose development began two years ago. It is based on the text-to-speech programming interface based on the Whisper model, designed for speech-to-text transcription and translation. Thus, this tool makes it possible to “generate natural voices that closely resemble those of the original speaker. “ It is remarkable that a small model with a single 15-second sample can create emotive and realistic vocals», We read on the OpenAI website.

Moreover, the company has published certain extracts and the result is truly astonishing. The audios obtained with Voice Engine have natural intonations close to those of the sample. For those wondering what this tool would be used for, the Californian company hopes to deploy it in order to:

• Provide reading assistance to non-readers and children;
• Translate content, such as videos and podcasts;
• Support non-verbal people;
• Helping patients regain their voice, for those suffering from sudden or degenerative speech disorders.

An innovation that worries

OpenAI’s intentions are laudable, but nothing prevents some from seeing this tool as a threat. We can’t blame them, because the risk of misusing the application for malicious purposes is increased. Indeed, voice cloning using AI tools has already proven its dangerousness and the usurpation of Joe Biden’s voice last January is proof of this.

As a reminder, during the Democratic primary vote held in January 2024, calls imitating the president’s voice encouraged voters not to vote. It later turned out that Steve Kramer, a consultant working for the campaign of Democrat Dean Phillips (a rival of Joe Biden), was behind the rigged calls. To this end, the FCC has decided to ban AI-generated calls in order to combat political and commercial fraud.

Aware of the risk that its tool can present, the Californian company is reassuring. “We collaborate with U.S. and international partners in government, media, entertainment, education, civil society and beyond to ensure we incorporate their feedback as we build» she wrote.

Furthermore, OpenAI specifies that these partners have accepted its usage policies “that prohibit impersonation of another person or organization without consent or legal right“.