Meta has announced the launch of AudioCraft, a new generative AI platform that allows users to create original music and audio content with just a few clicks. AudioCraft is powered by deep neural networks that can learn from a large corpus of music and audio data, and generate new sounds and compositions that match the user's preferences and specifications.
Users can access AudioCraft through a web interface or a mobile app, and choose from a variety of genres, moods, instruments, and effects. They can also upload their own audio samples or recordings, and use them as inputs for the generative AI.
AudioCraft can produce music and audio for various purposes, such as podcasts, videos, games, ads, or personal enjoyment. Users can also share their creations with others on the platform, or export them to other apps or devices.
Meta claims that AudioCraft is not intended to replace human musicians or composers, but rather to empower them with new tools and possibilities. The company also says that it respects the intellectual property rights of the original creators, and that it will not use or monetize the user-generated content without their consent.
Different Models of the AudioCraft AI Toolkit
Meta says the service consists of three models: MusicGen, AudioGen, and EnCodec:
- “MusicGen can generate music from text prompts. It was trained on Meta-owned and specifically licensed music, so it can create music that sounds professional and polished.
- AudioGen can generate audio from text prompts. It was trained on public sound effects, so it can create a wide variety of environmental sounds and sound effects.
- EnCodec is a decoder that allows for higher quality music generation with fewer artifacts. It is used by both MusicGen and AudioGen to improve the quality of their output.”
AudioCraft is currently in beta testing, and Meta plans to roll it out to more users in the coming months. The company also hopes to collaborate with artists, researchers, and developers to improve and expand the platform's capabilities and features. According to Meta, image and text generative AI solutions have become mainstream while audio models have taken a “backseat.”
The company also says that it will open source the models to give developers and researchers access to create their own solutions. However, a recent study shows Meta's definition of open source may be skewed, so it is unclear exactly how open AudioCraft is.
In a recent paper, a group of AI researchers from Radboud University in Nijmegen, Netherlands, argued that the term “open-source” is being used misleadingly by some companies. They point out that some generative AI LLMs that are labeled as “open-source” are not actually open-source, because the code that was used to train them is not available to the public.
Meta's Llama 2 and the GPT/codex model that underpins OpenAI's ChatGPT are two notable examples. The researchers argue that the lack of open-source LLMs is a problem for the AI community. They call on companies to release more open-source LLMs, so that researchers and developers can have access to the code and improve the performance of these models.