In a bid to catch up with its Big Tech peers in the AI arms race, Meta has been unveiling a slew of AI tools, the latest of which is Audiocraft, an AI tool that can generate audio and music from text prompts.
AudioCraft consists of three models: MusicGen, AudioGen and EnCodec. MusicGen was trained on roughly 400,000 recordings along with text description and metadata, amounting to 20,000 hours of music owned by Meta or licensed specifically for this purpose. It generates music from text prompts, while AudioGen, which was trained on public sound effects, generates audio from text prompts.
Today, Meta released an improved version of the EnCodec decoder, which allows higher-quality music generation. Simultaneously, the company is launching its pre-trained AudioGen models, enabling users to create an array of ambient sounds and auditory effects such as a dog’s bark, car horns, or footsteps on wooden surfaces. Additionally, Meta is making the complete set of AudioCraft model weights and code accessible to the public.
These models will be open-sourced, allowing researchers and practitioners to train their own models with their own datasets. According to Meta, The AudioCraft family of models is capable of delivering high-quality audio, while remaining user-friendly.
AudioCraft serves as a unified platform encompassing music, sound, compression, and generation, all within a single framework. Individuals aiming to build better sound generators, compression algorithms, or music generators can do so within the same code base, building upon the foundation laid by others in the field.
Source: mPost
