Meta Unveils AudioCraft, An AI Tool To Create Music And Audio From Text Prompts
Meta Platforms on Wednesday announced a new open-source artificial intelligence (AI) tool called AudioCraft that can create music and audio based on simple text prompts. Think of ChatGPT, but for creating music. Fascinating, right?
The AI tool is powered by three models, AudioGen, EnCodec, and MusicGen. According to Meta, AudioCraft works for music, sound, compression, and generation. MusicGen was trained by Meta using only company-owned and licensed music, the company claimed.
Creating music from scratch
Meta's blog post claims that MusicGen is able to generate music from text prompts while AudioGen creates audio from text prompts. In addition, Meta has also released an update to the EnCodec decoder that can help create high quality music.
The AudioGen models are pre-trained and are able to generate environmental sounds and effects like a dog's bark or sirens from vehicles. The model will be available to researchers to train their own models through their own datasets. According to Meta, the models are able to produce high-quality audio with long-term consistency and were developed at Meta over the last few years.
The company hopes that AudioCraft models would serve as helpful tools for musicians and sound designers in the future. Based on user feedback, Meta is also working on improving the current models and adding improvisations.
Meta isn't the first company to create a model like this one. Earlier this year, Alphabet (Google's parent company) introduced its own audio generation tool called MusicLM.
Even then, there are concerns surrounding such tools over copyright violation and the threat to musicians' jobs. Usually, data is scraped from the internet to train AI models without consent.