The AI tool is bundled with three models, AudioGen, EnCodec
and MusicGen, and works for music, sound, compression and generation, Meta
said.
MusicGen is trained using company-owned and specifically
licensed music, it added.
Artists and industry experts have raised concerns over
copyright violations, as machine learning software work by recognizing and
replicating patterns from data scraped from the web.
According to the company's blog post, MusicGen generates
music from text prompts, while AudioGen generates audio from text prompts. Meta
has also released an improved version of our EnCodec decoder. It helps in
generating higher quality music with fewer artifacts. The pre-trained AudioGen
models, announced by the company, lets users generate environmental sounds and
sound effects.It helps in generating sounds like a dog bark, or sirens from
vehicles.
The models will be available for access to researchers and
practitioners to train their models with their own datasets. These models are
capable of producing high-quality audio with long-term consistency, claims the
company. They have been internally developed at Meta over the past years.
AudioCraft models will act as tools for musicians and sound
designers in the future, the company said. The company is also working towards improving
the current models and add improvisations based on the users' feedback.
Earlier this year, Alphabet introduced its own experimental
audio generating AI tool called MusicLM.
0 comments:
Post a Comment