On Wednesday, Meta Platforms unveiled AudioCraft, an open-source AI application that enables users to produce music and audio based on text suggestions.
According to Meta, the AI tool works for music, sound, compression, and generation and is included with three models: AudioGen, EnCodec, and MusicGen.
The company’s own and specially licensed music is used for MusicGen’s training, it was added.
In light of the fact that machine learning software works by identifying and recreating patterns from data scraped from the web, artists and industry experts have expressed worries about copyright issues.
The company’s blog article states that while AudioGen creates audio from text prompts, MusicGen creates music from text prompts. Another updated version of our EnCodec decoder has been made available by Meta. It aids in producing music of greater quality and with fewer artifacts. Users can create environmental noises and sound effects using the company’s pre-trained AudioGen models.It facilitates the production of noises like a dog barking or car sirens.
Researchers and professionals will get access to the models so they can use their own datasets to train the models. According to the manufacturer, these devices can consistently provide high-quality audio. Over the past few years, Meta has been developing them internally.
According to the business, future musicians and sound designers will use AudioCraft models as tools. The business is also attempting to enhance the present models and incorporate improvisations in response to user feedback.
Alphabet unveiled MusicLM, an experimental AI tool for creating audio, earlier this year.