Meta introduced Audiobox - AI for voice generation and sound effects
Meta
Meta has announced a new AI platform called Audiobox, which allows you to create personalised voices and sound effects using voice commands.
Here's What We Know
Audiobox is based on Meta's earlier development, Voicebox. However, according to the developers, the novelty surpasses its predecessor in sound quality and functionality.
Starting today you can try our new foundation research model for audio generation. The demo includes Zero shot TTS, Text to sound effects, Infilling and more!
- AI at Meta (@AIatMeta) December 11, 2023
Try Audiobox ➡️ https://t.co/8OPcJYy8a9 pic.twitter.com/lo9rCOZMAh
Audiobox's main difference is the ability to not only generate but also edit audio. The platform can create speech in different languages, sound effects (car horns, dog barks, thunderclaps) and entire soundscapes.
Meta has built in Audiobox the means of control over the generation process, so that the desired result can be adjusted as precisely as possible.
To prevent abuse, the developers have integrated a digital watermarking system into Audiobox. Any audio generated using the platform is automatically labelled so that its origin can be traced.
According to Audiobox developers, the speed of audio generation is 25 times faster than previous Meta AI models. This is made possible by using personalised data processing algorithms for each specific task.
A limitation of the technology so far remains the lack of quality labelled data to train the AI model. For example, for Audiobox to accurately mimic different dog breeds or people's accents, it needs an appropriate set of examples. As the database expands, the platform's capabilities will only grow.
Source: Meta
