Meta’s Latest AudioSeal Tool Can Eliminate Voice Cloners Involved In Scams And Misinformation

Tech giant Meta says it has found the perfect solution designed to embed any hidden signals which are better known as watermarks seen in audio clips made by AI.

Facebook’s parent firm was quick to speak about this new feature that it claims is designed to assist in the detection of content made through AI online.

The endeavor called AudioSeal can be the company’s first feature that specifically highlights which parts in the audio clip were produced using AI technology. Similarly, the problems linked to the spread of misinformation or the uprising of scams via voice cloners can also be put to rest through this means, Meta explained.

The company’s top research scientist says the figures of malicious actors doing everything in their power to give rise to deepfakes of the country’s President has been an ongoing issue for a while now.

Scammers are using this technology to blackmail people and there is yet to be a solution that people can rely on. Now, however, things are changing as tech evolves and Meta is proud to introduce its own creation on this front that can tackle just that.

Furthermore, the scientist at Meta spoke in detail about how social media firms could detect and get rid of any unwanted content through this means as well.

Keeping all of this in mind, we do have a few caveats to mention. The company claims it has zero plans to apply the feature to AI-produced audio clips that are made through its own tools. Meanwhile, audio watermarks aren’t even adopted by the masses and so far, no industry standard is present for them.

Also, Meta’s experts shed light on how such content can be simple to play with. For instance, anyone can delete it or forge it with ease.

Now the question is how such a feature is going to be useful? Well, Meta says detecting fake or AI-generated content at a fast pace is a major benefit as is the chance to pinpoint elements across audio files produced using AI. All of this would be crucial in terms of ensuring the system is so useful, Meta added.

Today, Meta takes pride that 90 to 100% of all of its content is accurate in terms of watermark detection and better results would arise than that seen in the past at the watermarking audio.

For now, the feature is up for grabs through GitHub without any additional costs. If you wish to use it, simply download it and you can then add watermarks to any audio clip made using AI.

Also, users can overlay this across models produced through AI so it’s applied to all sorts of speech curated with this tech. But we do hope to hear more on this front as Meta’s experts will soon be presenting at this year’s International Conference on Machine Learning next month.

Read next:

• New Documents Reveal Elon Musk’s Plans To Store Money And Make Payments Through X

• Meta’s Latest API For Threads Enables Developers To Publish Posts, Fetch New Content, And Support Features Like Hiding Posts
Previous Post Next Post