This artificial intelligence, invented by Microsoft, can imitate any voice with a 3 second audio sample.
After AI capable of writing texts for masters and homework in any style, such as ChatGPT, a new tool from Microsoft can imitate the human voice. With a simple 3-second audio sample, the tool claims to be able to preserve both the tone and emotion of the original speaker for a more realistic rendering.
Trained with over 60,000 hours of English speech, this new Wall-E robot boasts military training and is the most advanced artificial intelligence on the market today, Microsoft says. Specialized in voice synthesis and text-to-speech, this tool primarily allows you to “synthesize high-quality personalized speech with just a 3-second recording of an unknown speaker as an acoustic reference.” An impressive learning capacity that raises many technical promises, but also certain ethical questions.
Even more realistic deepfakes
AI already knows how to imitate faces that are more real than life. With this new tool, she will soon be able to synthesize any voice, especially if it is a public person, whose image is logically more known on the network. Something that revolutionizes certain cinematic processes, like bringing some dead actors and actresses back to life. AI is no longer a miracle, as it already gave the impetus to actor Harrison Ford in the latest installment of Indiana Jones.
More worryingly, however, the technology behind Vall-E can also greatly improve the quality of deepfakes, allowing one face to be placed on top of another. By stealing the victim’s face and voice, it will be much more difficult for the general public to distinguish good from evil.
For now, Vall-E is only available in English and is not available to the general public. However, you can see the tool in action on a dedicated GitHub page with examples made for research purposes. The robot still has some drawbacks, especially in terms of audio transcription. Nevertheless, the result is impressive, especially since Microsoft is only at the beginning of its research.