Meta introduces AI-generated music, audio from text descriptions

• Meta and Microsoft collaborate on the new Llama 2 large language AI model.
• Meta develops generative AI tools for Instagram, including content identification.
• AudioCraft was released as open-source code.

Meta and Microsoft have recently collaborated to unveil the new Llama 2, a cutting-edge large language AI model, leading to Meta’s development of generative AI tools for Instagram, including one to detect AI-generated content.

Now, Meta has introduced its latest AI tool called AudioCraft in a recent blog post.

AudioCraft is an AI tool that generates high-quality and realistic audio and music from text descriptions. For instance, it could help small business owners add soundtracks to their video ads on Instagram effortlessly.

This tool promises to streamline the process of adding music to content creation, sparing users from the arduous task of searching for suitable songs.

Meta is not releasing AudioCraft on their platforms just yet, but they are making the tool’s code open-source.

This move allows researchers and practitioners to train their own models with custom datasets, contributing to the advancement of AI-generated audio and music.

Also Read

Meta CEO Addresses ‘Threads’ User Drop-Off and Future Plans

Zuckerberg addresses user drop-off on Threads, acknowledges need for improvements in retention....

AudioCraft is composed of three models: MusicGen, AudioGen, and an improved version of EnCodec. MusicGen is specialized in creating music and was trained on a vast dataset of 400,000 music recordings, accompanied by text descriptions and metadata.

AudioGen, on the other hand, generates lifelike environmental sounds based on written acoustic scene descriptions. The EnCodec decoder ensures higher-quality music generation with fewer issues.

Meta emphasizes the importance of responsible innovation and acknowledges that their training datasets lack diversity, especially in terms of music styles and language.

By sharing the code for AudioCraft, they hope to encourage other researchers to work on reducing biases and potential misuse in generative models.

Meta is excited to witness the creative outcomes that people will produce using this AI tool.

They have already shared hundreds of samples generated by AudioCraft, ranging from 80s disco and jazz instrumentals to people speaking with a background of enthusiastic cheering.