Meta’s three AI models can generate sound effects and music based on descriptions.
Meta, the internet corporation behind sites such as Facebook and Instagram, has released AudioCraft, its latest venture in the field of artificial intelligence. Through simple language prompts, this collection of generative AI tools enables content creators to create elaborate audio landscapes, melodies, and orchestral simulations.
The AudioCraft Components
AudioCraft is a versatile toolbox for music and audio production that consists of three core components:
AudioGen: This module generates a variety of audio effects and immersive soundscapes, allowing you to create genuine aural experiences such as a barking dog, a honking automobile horn, or footfall on varied surfaces.
MusicGen: Designed to create musical compositions and melodies based on verbal descriptions, MusicGen produces tunes with catchy melodies, tropical rhythms, and more.
EnCodec: EnCodec, an innovative neural network-based audio compression codec, improves music production while decreasing artifacts, providing a refined audio experience.
Taking Care of the Gap in Generative Audio Development
Meta recognizes that generative AI models focused on text and graphics have gained traction, whereas audio tools have progressed more slowly. The MIT License distribution of AudioCraft aims to bridge this gap by providing accessible tools for audio and musical exploration, supporting innovation in this relatively underdeveloped domain.
Towards a More Comprehensive Community Contribution
The goal of Meta is to enable academics and practitioners to investigate, experiment, and construct their own models using AudioCraft, which coincides with the company’s commitment to pushing the boundaries of generative music. The availability of these AI models serves as the foundation for a collaborative approach, allowing the larger community to contribute to the advancement of audio synthesis.
Development and Ethical Considerations
Unlike several image synthesis models that have received criticism for using hidden or unethical training material, Meta highlights MusicGen’s ethical foundation, stating that it was trained on a collection of 20,000 hours of music owned or licensed explicitly for this purpose. This ethical perspective may be appealing to skeptics of generative AI.
Generative Audio Has a Bright Future
The integration of AudioCraft into the open-source ecosystem by Meta has the potential to spur innovation and creativity in the field of generative music. Developers and fans alike will be able to use these tools for a variety of purposes, perhaps leading to the birth of user-friendly and novel generative audio solutions.
