Meta's Audiobox: Innovative Audio Generation for Diverse Uses

Article by: Manish Methani

Last Updated: December 12, 2023 at 10:08am IST
2 min 9 sec read

Meta's Audiobox, the successor to Voicebox, is a groundbreaking model designed to revolutionize audio generation. Here's what you need to know about this cutting-edge technology:

Key Features of Audiobox:

• A new research model for audio generation, expanding on Voicebox's capabilities.
• Employs voice inputs and natural language prompts to create custom audio, including speech, sound effects, and soundscapes.
• Offers enhanced controllability, allowing users to describe desired sounds or voices using simple text prompts.

Audiobox's Capabilities:

• Generates various audio types, such as speech in different environments, styles, and non-speech sound effects.
• Allows for voice and text inputs, providing precise control over the generated audio's style and content.
• Showcases superior performance in quality and relevance compared to prior models, outperforming Voicebox in style similarity by over 30%.

Responsible Research Collaboration:

• Invites researchers and academic institutions to apply for grants, focusing on safety and responsibility research with Audiobox.
• Released under a research-only license to selected partners, aiming to lower barriers for audio content creation.

Safety Measures:

• Implements audio watermarking for traceability and includes voice authentication to prevent impersonation.
• Tested across diverse speakers and languages for consistent performance and robustness.

Future Prospects:

• Aims to evolve into a generalized audio generative model, enabling a wide range of creative uses.
• Envisions a future where Audiobox facilitates diverse audio creation for both professionals and hobbyists, benefiting industries like gaming, content creation, and AI development.

Meta's Audiobox heralds a new era in accessible and versatile audio generation, promising innovative applications across various fields.

You can read the full blog post and see demos on official Meta AI blog post here.