Meta Unveils Llama 3.2—AI Now Talks Back with a Twist!

“Meta Releases Llama 3.2—and Gives Its AI a Voice”

“Meta Platforms, the company behind Facebook, Instagram, and other social platforms, recently unveiled an artificial intelligence model that sounds unnervingly like a real human. The company said this achievement was the result of a new approach to teaching AI systems how to decode and reproduce human speech. Its engineers used a combination of machine learning techniques and enormous datasets to train the model, dubbed as Voice ID.”

Oh, cue the sci-fi movie soundtracks and get ready to toss your Alexa out the window. Meta Platforms Inc., the social network conglomerate we’re all so acquainted with, just added another feather to its AI cap. Upstaging their own party, they’ve rolled out an AI model, charmingly christened ‘Voice ID’, sounding alarmingly like a genuine human.

Much to our amazement, this feat, they claim, is an outcome of a novelties in making AI systems decipher and mimic human speech. You guessed it right! An amalgamation of advanced machine learning methods and mammoth datasets was the secret recipe for creating this unsettlingly realistic AI voice. It’s like the perfect blend of Frankenstein’s monster and HAL from “2001: A Space Odyssey”.

The real bite is that this model can mimic anyone. Yes, you read it right – anyone. Forget Mary had a little lamb, it’s more about Meta had a creepy Llama. All it needs is a bare minimum of five minutes of one’s speech, frighteningly less if you ask me, to generate a synthetic version of the individual’s voice. Compellingly real enough to make you wonder whether you’re talking to John Doe or his creepy Voice ID twin.

Quite fascinating is the usage of ‘Transformer Sequence-to-Sequence’ (Transformer TTS) system to train this wonder kid. The raw charm of sentence level transcript justified by the word ‘Transformer’ here, taking the AI game a notch higher. And here’s a bit of a spoiler, it not just mimics the voice but the beats and the tones, making it sound way more realistic.

This development, while undeniably a signal of the ever-advancing AI field, is not without concern for those interested in privacy and ethical implications. Suddenly, ‘Is that really you talking?’ doesn’t seem like such a funny question anymore. The fear of voice phishing just got real, friends. Remember, always think twice before saying ‘Hello’ to that eerily familiar voice on the other side of the line.

Finally, folks, be aware. While this AI could be the future of personalised voice assistants or a pioneer in making audio content more accessible, don’t forget, what goes up must come down. As sophistication climbs, the fall might just be that much harder if we trip over ethical boundaries. So, let’s tread with vigilance into this brave new world of unsettlingly-real AI voices.

Read the original article here: https://www.wired.com/story/meta-releases-new-llama-model-ai-voice/