In the digital age, artificial intelligence (AI) is becoming more pervasive, finding applications in various sectors. Recently, we have seen the progress of Google's project to generate music from a simple 'prompt'. Today, we find that too Meta embarked on this adventure, creating language models capable of generate audio from a text description. The name of this revolutionary tool is 'audio craft'.
Meta Audiocraft: a new horizon for the generation of music available to everyone (or almost)
Audiocraft is an AI-based tool that makes use of three language models: AudioGen, EnCodec and MusicGen. The goal is to allow users to obtain audio tracks starting from a simple text description of what they want to hear. Audiocraft can be used to generate both music and sound, as well as offering compression capabilities.
Among the three models, MusicGen is the one specifically trained to generate music from text. To do this, thousands of musical pieces licensed by Meta were used. The most exciting news is that all three models have been released in open source, which means that anyone can use them to create tools and applications. According to Meta, these templates were specifically designed for music artists and sound designers, with the aim of "providing inspiration."
Read also: Meta's new strategy for targeted advertising: user permission
The templates are already available and their code can be downloaded through hugging face or the related page Github. MusicGen can be used to get AI-generated song snippets, while AudioGen can provide usable sounds for music production. EnCodec, on the other hand, is an AI-based compression tool.
To generate a piece of music, that's enough describe the music you want to get and press the 'Generate' button. Furthermore, it is possible to accompany the description with an audio file or a recording of your microphone to allow the model to more accurately process the type of music you want to generate.
The audio generated will be a track approximately 12 seconds long. To get a tool that can generate more complete snippets, we'll have to wait for some developer to build an app based on these templates. Meta provided tools, now it's up to the community to find out what can be done with them.