A new open-source Al tool called AudioCraft has been made available by Meta. With the help of text prompts, users of this generative Al tool can compose music.
The program combines the AudioGen, EnCodec, and MusicGen generative Al models.
The tool MusicGen creates music from text inputs. Over 20,000 hours of music that Meta either owns or has obtained a license for were used to train the system. Users can produce sounds with fewer artifacts because of Meta’s EnCodec decoder, which also prevents distortion from audio manipulation. Contrarily, AudioGen generates sounds based on written instructions, such as imitating the sound of dogs barking or footsteps. It received training using common sound effects.
While Al-produced photos and text have become more popular, sound has not fully caught up yet, according to Meta. The company claims that previous sound projects have been difficult to access and complicated. The company is also open-sourcing these models to researchers, allowing them to train these models as their own using their datasets. The new toolkit intends to empower developers to personalize their models and push the frontiers of what is possible.
“AudioCraft works for music, sound, compression, and generation – all in the same place,” wrote Meta in a blog post introducing the tool. The manufacturer claims that the simplicity of usage and reusability of this AudioCraft are two of its benefits. Therefore, by building on what others have already accomplished, those interested in creating better sound generators, compression algorithms, or music generators can use this tool and improve it.
However, because it requires technical expertise to utilize the tool successfully, AudioCraft is not designed for average users. According to the company, its main use is for research. The performance and control strategies of these models are currently being improved by the developers to increase their functionality.
Google recently unveiled MusicLM, another text-to-music technology that functions similarly and generates music from word prompts.
