OpenAI previews ability of Voice Engine to convincingly clone a person’s voice with 15 second voice sample

OpenAI has previewed the current state of its Voice Engine technology that can convincingly clone a person’s voice using a 15 second voice sample as input. The technology can also transfer a person’s accent into other spoken languages while translating speech, even if the target language uses informal, or slang, speech. For speakers with voice impairments or illnesses that result in unclear speech, like laryngitis, Voice Engine can repeat what is said in a clear voice.

AI technology has advanced to the point where it recognizes vowels, words, and other parts of speech and can understand the gist of sentences. Voice cloning AI recognizes the unique traits of a person’s speech, such as accent, emotion, timing, and emphasis, then uses those characteristics to speak text as a convincing clone.

OpenAI demonstrated on its blog page convincing examples of:

• Voice cloning
• Speech translation with voice accent cloning
• Speaking informally, or in slang
• Speaking for the mute
• When suffering from speech conditions, speaking in a person’s original, clear voice

OpenAI is not releasing the Voice Engine to the public at this time due to concerns of misuse, despite many other AI voice cloning and voice adaptation services on the market. Such technology has already been used during the US election cycle to create ‘fake President Biden’ phone calls, and across the world to scam money from companies and people. Unfortunately, once Pandora’s box has been opened, like generative AI image technology used to create fake Pope images, there’s no turning back.

Concerned readers should create safe words with family members and close friends to verify their identities, read how to recognize scam calls, disable the use of voice recognition verification with financial providers, and consider using a voice changer to protect against having their voice copied when answering unknown callers.