Meta has launched the demo of their newest generative artificial intelligence (AI) called, Audiobox
It comes with the tagline ‘Where anyone can make a sound with an idea’ and this new AI is built on, and replaces, Voicebox, their previous voice-generating AI.
According to a blog post by the social media giant, Audiobox is an improvement because it combines voice generation with sound effects and soundscape generation, vastly increasing the possibilities. The tool’s webpage states “It can generate voices and sound effects using a combination of voice inputs and natural language text prompts — making it easy to create custom audio for a wide range of use cases.”
Starting today you can try our new foundation research model for audio generation. The demo includes Zero shot TTS, Text to sound effects, Infilling and more!
Try Audiobox ➡️ https://t.co/8OPcJYy8a9 pic.twitter.com/lo9rCOZMAh
— AI at Meta (@AIatMeta) December 11, 2023
It’s as simple as typing in a sentence you want the AI to say and choosing what kind of voice you want to hear it in, or a sound effect you want it to create, and Audiobox will generate it for you. But you can also go one level deeper and record your voice and have the tool speak with a generated version of it. You only need to record a sentence or two to get an uncannily accurate result.
Users can experiment with the tool now, but Meta makes it clear that the current release is a research demo only, and “may not be used for any commercial purpose(s).” It also stipulates that residents of Texas and Illinois cannot use the tool. VentureBeat reports that this is due to state laws preventing the kind of audio data collection required for the demo to function.
Concerns about AI and society
Earlier this year the owners of Facebook disbanded their “responsible AI” team in favor of investing more heavily into generative AI. Along with the recent expansion of their AI image generator into a standalone tool, the launch of Audiobox is a clear step in that direction. Neither of these new tools is open source, which is a surprising move after Meta’s release of Llama 2, an open-source competitor to ChatGPT and other large language models.
There is no information in the blog post, or the research paper that was released alongside the demo, as to the nature of the data collection undertaken for this model. This will undoubtedly raise eyebrows as the world scrambles to understand how AI models fit into copyright laws.
Featured image: The Digital Artist/Pixabay