AI Startup Can Now Generate Speech Using Your Voice In 30 Languages

The AI arms race continues to speed up, with new frontiers in voice cloning rising virtually every day. The newest growth comes from San Francisco-based startup ElevenLabs, which simply introduced that their new AI mannequin can now mimic voices talking fluently in 30 completely different languages—a dramatic enlargement from the unique eight that had been beforehand supported.

The corporate used Lukeman Literary, a literary company and impartial writer, for example, explaining that the corporate produces many audiobooks every year in a number of languages.

“It used to take Lukeman’s group weeks to provide a single audiobook as a result of it required them to seek out the appropriate voiceover artist, ebook a recording studio, and report and handle the post-production,” ElevenLabs mentioned in an official weblog submit. “ Now your complete course of takes a couple of hours,”

In line with ElevenLabs, the brand new Multilingual v2 mannequin delivers “emotionally wealthy” audio that captures the nuanced inflections of pure speech. Customers sort the textual content they need spoken within the goal language, and the AI generates a seamless voiceover.

The corporate supplies two important voice cloning choices: a text-to-speech software and a “VoiceLab” for cloning particular voices.

Customers add speech samples to create a customized voice clone, which the AI analyzes to construct an artificial model. This cloned voice can then be manipulated to say something possible. ElevenLabs claims the most recent replace means these AI doppelgangers can now communicate fluently in tongues like Swedish, Arabic, and Malay.

The expanded linguistic capabilities additionally coincide with ElevenLabs transferring its voice cloning tech out of beta testing. The corporate goals to market the software for sensible functions like narrating audiobooks, as within the case of Lukeman Literary.

Addressing considerations

The expertise’s potential for misuse clouds these enterprise ambitions. Deepfake audio leaves customers weak to fraud and misinformation campaigns. ElevenLabs itself endured backlash final yr when its platform was exploited to impersonate and harass public figures.

The corporate says extra stringent safeguards have since been applied, however moral considerations persist. As Decrypt just lately reported, a “scammer may use AI to clone the voice of your beloved,” and all it might require to realize plausible outcomes are a few minutes of audio.

Main tech corporations like Meta face related criticism for growing highly effective generative AI with out full transparency. Meta just lately unveiled an AI speech synthesis software referred to as Voicebox, which it acknowledged may simply facilitate deepfakes. Not like ElevenLabs, Meta kept away from any public launch given the “dangers of misuse.”

Nevertheless, regardless of the fears, speedy progress in AI voice cloning appears unstoppable. As linguist Mati Staniszewski of ElevenLabs said, “Finally we hope to cowl much more languages and voices with assist of AI and get rid of the linguistic obstacles to content material.”

Making certain moral implementation stays a steep problem, as the road between world misinformation and progressive methods to speak may be very skinny. Treading fastidiously is vital—lest our world village of voices turns into a cacophonous Tower of Babel.