Meta unveils AI tool for music generation

The company Meta has developed a generative artificial intelligence (AI) AudioCraft, which creates sounds and music from text descriptions.

The neural network is capable of generating environmental sounds — for example, a dog’s bark or footsteps — and also processing sounds from musical instruments.

“Imagine a professional musician who can create new compositions without playing a single note on an instrument. Or a small-business owner who easily adds a soundtrack to their latest Instagram post,” the company described the technology.

AudioCraft consists of three models:

• MusicGen — a neural network trained to generate music from text prompts;
• AudioGen — a tool that creates sound effects;
• EnCodec — an AI decoder that improves recording quality.

Meta has released the source code of the models on GuitHub. At the same time, the company uses an already licensed library of sounds and music that does not infringe copyright.

The company expects AudioCraft to be used by sound engineers and other content creators.

Earlier in August, the Financial Times reported on Meta’s plans to launch a chatbot with support for multiple personalities. The AI was slated to appear in September 2023.

For the April–June period, the operating loss of the metaverse-related Reality Labs unit within the corporation stood at $3.74 billion.