Meta, the social media giant formerly known as Facebook, has launched a new suite of AI models called AudioCraft, which generates music and audio based on text prompts. This development sets up a direct competition with Google’s MusicLM, a text-to-music generator that was introduced a few months ago.
The AudioCraft suite consists of three models: MusicGen, AudioGen, and EnCodec. These models allow users to create music and sound effects by simply providing text prompts such as soulful music for a dinner party or movie scene in a desert with percussion. According to Meta, AudioCraft is like a new type of instrument akin to the advent of synthesizers.
MusicGen, the model responsible for generating music, was trained on 20,000 hours of Meta-owned and licensed music. However, it remains unclear whether EnCodec, which produces higher quality music, was trained on copyrighted material or adhered to the same guidelines as MusicGen. Meta has not yet responded to inquiries regarding this matter.
Training AI models is a contentious issue within the industry. To achieve human-like outputs, these models often train on vast amounts of data, including copyrighted material, without proper authorization, compensation, or knowledge of the copyright owners. It is a practice that many of the leading AI companies engage in.
In an effort to promote transparency and tackle bias, Meta has made all three models—MusicGen, AudioGen, and EnCodec—available as open-source models. This enables researchers and practitioners to access and train their own models using their own datasets. Meta hopes this will further advance the AudioCraft tools and address any biases, especially Western-style music, which currently dominates its training set.
Creating music with AI is a complex task as it involves capturing both local and long-range patterns, from individual notes to the overall musical structure with multiple instruments. However, Meta claims that its suite of models is capable of producing high-quality audio with consistency and ease of use.
This move by Meta reflects the increasing interest and investment in AI-generated music and audio. It opens up new possibilities for musicians, content creators, and music enthusiasts who can now experiment with sound by simply providing text prompts. However, concerns surrounding copyright and licensing issues persist within the industry.
As the AI industry continues to evolve, it is crucial to strike a balance between innovation and respecting intellectual property rights. Meta’s launch of AudioCraft contributes to the ongoing conversation surrounding the ethical use of AI and the need for clear guidelines and regulations. With open-source availability, the potential for further advancements in AI-generated music and audio is within reach.
Overall, Meta’s AudioCraft suite presents an exciting development in the field of AI-generated music and audio, offering users a new and accessible tool for creative expression. As the competition with Google’s MusicLM heats up, we can expect even more innovations and advancements in this space.
