Meta's Fundamental AI Research (FAIR) team has announced the release of several new AI models and tools aimed at fostering advancements in AI research, particularly in audio generation, text-to-vision capabilities, and watermarking technologies.
In a press release, Meta expressed its commitment to fostering innovation by sharing early research work publicly, aiming to inspire iterative improvements and responsible AI development.
Meta introduces JASCO, a new AI model designed for text-to-music generation. JASCO allows users to input various audio parameters, such as chords or beats, to refine and control the generated music output. According to FAIR researchers, this model empowers users to adjust features like chords, drums, and melodies through text inputs, enhancing the precision and customization of AI-generated music.
Meta plans to release JASCO's inference code as part of its AudioCraft AI audio model library under the MIT license, with the pre-trained model available under a non-commercial Creative Commons license.
Meta introduces AudioSeal, a novel watermarking technique tailored for detecting AI-generated speech within longer audio segments. This tool marks a significant advancement in content identification, enabling localized detection of AI-generated content with increased speed and efficiency. Meta claims AudioSeal enhances detection speed by 485 times compared to existing methods. Unlike other models, AudioSeal will be available under a commercial license.
FAIR is releasing two variants of its Chameleon multimodal text model, Chameleon 7B and 34B, under a research-only license. These models integrate textual and visual understanding capabilities, supporting tasks such as image captioning. However, Meta clarified that the Chameleon image generation model will not be released at this time, focusing solely on text-related functionalities.
Additionally, Meta will provide researchers access to its multi-token prediction approach, which trains language models to predict multiple future words simultaneously, rather than sequentially. This approach will be accessible under a non-commercial and research-only license.
Meta's latest releases underscore its dedication to advancing AI research across diverse domains, empowering researchers to explore and innovate in AI-driven technologies responsibly. These initiatives are poised to contribute significantly to the evolution of AI applications in various fields, from multimedia content generation to enhanced data security and processing efficiency.
