Not planning to leave audio behind in the world of AI, Meta has
launched a new AI project ‘AudioCraft,’ designed for people to create “high
quality audio and music” using text prompts. Does this mean that anyone can now
produce music without playing notes, instruments, or specific skills? Seems so.
AudioCraft is created based on open-sourced research models of
generative AI including MusicGen, AudioGen and EnCodec. The characteristics of
the three models are as follows:
MusicGen is capable of generating music from text prompts. It
has been trained with Meta-owned, licensed music.
AudioGen has been trained on public sound effects including environmental
sounds and sounds of nature and animals. Like MusicGen, it too generates audio
from text prompts.
EnCodec is a decoder, that Meta has just released a revamped version of. This model generates higher quality music with fewer prompts.
By open-sourcing these models, Meta is providing researchers
and practitioners the opportunity to train their own models with their own
datasets, while also contributing to the field of AI-generated audio and music.
Explaining the complexity of generative AI systems for
audio, Meta says that “generating high-fidelity audio of any kind requires modelling
complex signals and patterns at varying scales.” However, as compared to
previous AI audio projects, AudioCraft models are simplified and easy to use.
While AudioCraft points to a fascinating future for the world
of music composition, like other AI models, it too will have unpleasant consequences.
AI-generated audio can easily cause disruptions in the music industry, as we
have already seen happening with music tracks featuring recording artists
generated purely by AI, with no actual involvement of the artists.
Moreover, music is essentially a form of art and to master
it requires adequate skills. A mere integration of AI could take music
composers, with no real knowledge and understanding of music production, only
to a superficial level of achievement. In other words, actual human talent and
skills can not really be replaced by digital systems. Having said that, generative
AI could still be a fun and creative way to experiment and play around with audio
and music.
🎵 Today we’re sharing details about AudioCraft, a family of generative AI models that lets you easily generate high-quality audio and music from text.https://t.co/04XAq4rlap pic.twitter.com/JreMIBGbTF
— Meta Newsroom (@MetaNewsroom) August 2, 2023
