We know that artificial intelligence has been developing and gaining popularity in recent times. Systems that create images of texts like Midjourney, DALL-E and models like ChatGPT that respond to everything we ask made an impression all over the world. Now is from Microsoft A brand new move in artificial intelligence has arrived.

US tech giant, artificial intelligence model that can create voice from text’VAL–UNTIL’introduced the The system, which can revolutionize artificial intelligence, can easily translate human voices. can imitate expressed. Of course, this kind of technology also brought some concerns.

Can imitate sounds with just a 3 second sample

According to Ars Technica, VALL-E just a three second audio clip It can imitate human voice. It is even said that what it can do is not limited to this, artificial intelligence can even produce results that match the tone of voice according to the emotion of the speaker.

Microsoft has announced that VALL-E, a language model, will be introduced by Meta in October 2022.encode’ He states that he has benefited from the so-called technology. Unlike similar systems we normally see, the model draws conclusions from text and sounds. Basically, how a person sounds is analyzingThanks to EnCodec, it divides this information into separate components and matches it with the training data. This creates different sentences by mimicking the sound in the example.

In a shared article on artificial intelligence, researchers used VALL-E, more than 7,000 of the speaker 60,000 hours of English He states that he trained with audio recordings in his language. It is said that for the system to produce a good result, the sound in the samples must be close to the sound in the training data.

Microsoft has released some samples of VALL-E on GitHub. When looking at the examples, it turns out that in some places artificial intelligence appears with the voice of a robot, but in others it is surprisingly surprising. realistic it looks like it. Also in the examples, VALL-E retains the speaker’s tone; even result by environment can also be seen. For example, if the original speaker speaks from a reverberant place, the system produces sound accordingly.

This kind of technology is not without risks.

Of course, this kind of technology is somewhat alarming. Malicious people can make it look like they said something they didn’t, can imitate and can lead to an increase in incidents such as fraud. You can think of it as the risks of deepfake, which has become popular lately. Open source code from Microsoft due to risks impossible However, we can say that similar technologies can carry these risks.