AudioCraft by Meta, a generative AI tool – Music from Text

Source: Meta

Meta Platforms has recently launched an open-source AI tool named AudioCraft, designed to enable users to generate music and audio content using text prompts. The tool includes three models: AudioGen, EnCodec, and MusicGen, catering to music, sound, compression, and content generation, as emphasized by Meta.

MusicGen, one of the bundled models, is trained using proprietary music from the company and licensed sources. However, concerns have been raised by artists and industry experts about potential copyright violations due to machine learning software relying on patterns recognized and replicated from web-scraped data.

Earlier this year, Alphabet Inc also introduced its experimental AI tool, MusicLM, for audio generation purposes.

In a forward leap into the realm of AI, Meta has introduced a cutting-edge AI tool called AudioCraft, designed to produce music and audio content based on text prompts.

All about AudioCraft

According to Meta, this innovative tool opens up exciting possibilities for professional musicians, allowing them to explore new compositions effortlessly without touching a single instrument. It also empowers small business owners to effortlessly add captivating soundtracks to their video ads on platforms like Instagram.

Meta envisions AudioCraft as a powerful AI tool that generates high-quality, lifelike audio and music from simple text inputs. The company acknowledges the remarkable progress made in generative AI for images, text, and video but acknowledges that the audio domain has been somewhat left behind. With the aim of bridging this gap, Meta has developed AudioCraft, offering a user-friendly platform that makes audio generation a breeze.

In their blog post, Meta highlights the challenge of the audio domain lagging behind other aspects of generative AI. While progress has been made, existing solutions are often complex and not easily accessible to the general public. AudioCraft aims to change this landscape and bring the power of generative audio to a wider audience.

The AudioCraft platform is built upon three distinct AI models: AudioGen, MusicGen, and EnCodec, each serving specific purposes.

AudioGen draws from a vast array of public sound effects and can generate diverse audio content based on text inputs, ranging from the blaring of horns to the barking of dogs and the sound of footsteps.

Summary

On the other hand, MusicGen is trained on Meta’s proprietary and licensed music, enabling it to craft original music compositions from the provided text prompts.