Meta Unveils AudioCraft: An AI Tool Transforming Text into Audio and Music.

Meta, the tech giant, has recently unveiled a groundbreaking open-source AI tool called AudioCraft. This innovative tool is intended to empower both professional musicians and everyday users to effortlessly create audio and music from simple text prompts.

AudioCraft consists of three distinct models: MusicGen, AudioGen, and EnCodec. MusicGen has been trained using Meta’s exclusive music library, allowing it to generate music based on text inputs. Conversely, AudioGen is trained on public sound effects data, enabling it to create audio from text inputs. Additionally, the EnCodec decoder has been enhanced to deliver higher-quality music generation with fewer unwanted artifacts.


The company is generously providing access to their pre-trained AudioGen models, enabling users to generate environmental sounds and sound effects like dogs barking, cars honking, or footsteps on a wooden floor. Furthermore, Meta is openly sharing all the model weights and code for the AudioCraft tool, which has vast applications, including music composition, sound effects generation, compression algorithms, and audio generation.

The decision to open-source these models is aimed at granting researchers and practitioners the opportunity to train their own models using personalized datasets, fostering creativity and advancements in the field.

Meta acknowledges that generative AI has made remarkable strides in images, video, and text, but the same level of development has been lacking in audio. AudioCraft sets out to bridge this gap, providing a more accessible and user-friendly platform for generating top-tier audio.


In their official blog post, Meta highlights the unique challenges in creating realistic and high-fidelity audio, as it involves modeling complex signals and patterns at various scales. Music, being a composition of local and long-range patterns, presents a particularly challenging aspect in audio generation.

AudioCraft is a game-changer in this regard, boasting the ability to produce high-quality audio over extended durations. The tool simplifies the design of generative models for audio, making it easier for users to experiment and explore the vast possibilities of existing models.

Meta’s release of AudioCraft marks a significant step forward in the evolution of AI-generated audio and opens up exciting new horizons for music enthusiasts, sound designers, and researchers alike.

Leave a Reply

Your email address will not be published. Required fields are marked *