Meta launches AudioCraft: a suite of generative AI models for audio and music creation

Meta introduces new AI tool that effortlessly generates top-notch audio and music from simple text prompts. This raises new concerns for patent infringement.

Electronics, Speaker, Computer Hardware, Hardware, Monitor, Screen, Adult, Female, Person, Woman, Headphones, Head, Earbuds, Clothing, Hat, Face, Accessories, Jewelry, Ring

Meta recently launched a new AI tool that transforms the landscape of audio and music production. AudioCraft comprises a plethora of generative AI models—MusicGen, AudioGen, and EnCodec—enabling users to effortlessly create top-tier audio and music from textual cues.

In contrast to the predominant emphasis on generative AI for images and text, the progression of audio has been comparatively sluggish due to its intricate complexity. MusicGen, which underwent training with licensed music, crafts musical compositions from text inputs. On the other hand, AudioGen, utilising publicly available sound effects, generates corresponding audio outputs. The enhanced EnCodec decoder further refines the quality of generated music, minimising unwanted artifacts. Additionally, Meta releases pre-trained AudioGen models that facilitate the creation of ambient sounds like dog barks, car horns, and footstep echoes on diverse surfaces. The sharing of model weights and code fosters collaborative open-source efforts, propelling AI-generated audio and music to new heights.

Meta envisions AudioCraft as a catalyst for igniting creativity among musicians and sound designers. This launch signifies a significant advancement in reshaping the creation and consumption of audio, inviting users to embark on an expedition into the unexplored potential of AI-crafted auditory landscapes.

Why does this matter? The rise of unauthorised content creation poses a growing challenge due to AI algorithms generating substantial content that might violate existing copyrights. The complexity of correctly identifying creators and owners adds to the intricacies of protecting and enforcing intellectual property rights. Establishing patent eligibility necessitates elements like uniqueness and non-obviousness. With AI now generating innovative inventions and findings, determining if an AI-produced invention meets patent criteria becomes complex, involving questions about attributing inventiveness to human involvement or AI.

Precisely for these concerns, the Universal Music Group (UMG), the world’s largest music rights company, has recently approached the US Congress for assistance in tackling copyright infringement related to emerging AI technology. This right would safeguard individuals’ likenesses, encompassing their names and voices, on a nationwide level. UMG also advocated for copyright owners to be informed about AI training models and for clear labelling of “AI-generated content.”