StabilityAI Unveils Stable Audio 2.0: Longer Tracks, Audio-to-Audio Support, and Copyright Protection

StabilityAI Unveils Stable Audio 2.0: Longer Tracks, Audio-to-Audio Support, and Enhanced Copyright Protection

StabilityAI has recently launched the second version of its artificial intelligence music generation tool, Stable Audio 2.0. This new iteration offers users the ability to create three-minute tracks at 44.1 kHz stereo by simply inputting a natural language processing prompt. The AI-generated tracks now include structured compositions with intros, developments, outros, and stereo sound effects, providing a more immersive music creation experience.

One of the key features of Stable Audio 2.0 is the introduction of audio-to-audio support, allowing users to upload audio files to the platform to generate fully produced samples. This evolution from text-to-audio tool opens up new possibilities for creators to experiment with different sounds and styles. For example, users can mimic a drum sound with their voice, prompting the app to create an audio clip of a drum playing.

In light of the new audio-to-audio feature, StabilityAI has reinforced its commitment to protecting the copyright of creators. The platform uses content recognition technology to ensure compliance with its terms and conditions, preventing any copyright infringement. Additionally, the AI model is trained on AudioSparx’s extensive audio file library, with musicians having the option to opt out if they do not wish their works to be used for AI model training.

The release of Stable Audio 2.0 comes after the departure of former VP of audio, Ed Newton-Rex, who resigned from his role citing disagreements with the company’s approach to copyright and creator rights. His departure has sparked a conversation within the tech industry about the ethical implications of using copyrighted works for AI training.

Under the hood, Stable Audio 2.0 features a new architecture that enhances the generation of full tracks with coherent structures. The tool utilizes a compressed autoencoder and a diffusion transformer to recognize and reproduce large-scale structures essential for high-quality musical compositions. StabilityAI has made the tool free to use and available immediately, inviting creators to explore the possibilities of AI-generated music.