Meta He has managed to amaze everyone with his project Audiobox, a generative artificial intelligence capable of reproducing a human voice from a few seconds of audio. It is not one of those projects that look very good on paper and then come to nothing, because the announcement has been accompanied by a public demonstration of these capabilities.
In addition to voices, Audiobox can also generate unique sounds through voice or text prompts. In this post we tell you everything that is known so far about this project and, best of all, how you can try it yourself.
What is Audiobox?
Audiobox is the name chosen to designate Meta's fundamental search model for audio generation. The creation of personalized audio, which can be applied in various situations and scenarios, is the result of combining voice input and text prompts in natural language.
As Meta explained through a statement, this is the fruit of many efforts and years of research. And only the first stone of a whole new world of possibilities that opens before us.
The importance of this launch must not only be assessed in the "what", but in who is behind the project. We must not forget that Meta is the company that controls some of the most used apps in Spain, such as Facebook, Instagram or WhatsApp. This opens the door to seeing implementations of this new technology in them in a not-too-long period of time.
What possibilities does Meta Audiobox AI offer?
Audiobox brings us six unique functions based on Artificial Intelligence for audio creation and editing. These put a wide range of customization options at our disposal. They are the following:
- Create audio with our own voice, based on any short audio sample, even a few seconds long. This function allows us to create a speech that imitates the tone and style of our own voice or that of another person.
- Voices described. The audio is generated from a series of guidelines described in a text. The best thing about this is that it makes it possible to create new and unique voices.
- Redesigned voices. The idea is to change the tone and style of a real voice using a text description. We could say that it is a combination of the two previous functions in favor of an even higher level of customization.
- Sound effects. In addition to voices, Audiobox by Meta is capable of generating sound effects from descriptive text.
- magic audio editor, a handy tool to remove annoying background noise from voice recordings.
- sound fill. A function through which to replace part of an audio with new sounds.
As you see, Audiobox offers many possibilities for audio professionals and content creators, although it is also very interesting for any curious user. For now, the voice actors can rest assured, since in view of the results, The voices generated by this AI are still a bit robotic, devoid of naturalness. However, it is a matter of time before these small inconveniences are overcome.
How to try Audiobox
The best way to test this new technology is to try it ourselves. This is possible through the web demo Audiobox, created recently, and still available completely free of charge. The way to try it is this: record our own voice (or play any other) and start rehearsing with it. This is just one of the possibilities that this technology offers us.
Although this testing page is available in Spain, at the moment can only be used in English, That is the language that we will have to use to request texts and generate audios. We have tried one of the female voices available with the phrase «This is a voice test for the web Movilforum» and this has been the result:
Meta's Audiobox Misuse and Other Worrying Issues
One of the most surprising features of Meta Audiobox is the ability to generate our own voice through this AI tool. But, at the same time, it also generates many doubts and uncertainties, because about it plans the threat of possible misuse.
In order to prevent this technology from being used to commit fraud or scams, Meta requires acceptance of a number of terms of use before allowing us to test this functionality.
Apart from this, the generated audios have a kind of "watermark" that allows their origin to be precisely traced. In the press release we mentioned before, Meta explains that this distinctive is actually a signal that is imperceptible to the human ear, but can be detected.