Stability AI Launches Music Generation Tool Stable Studio

2023-09-14

Stability AI, a company renowned for its AI-generated visual effects, has recently launched a text-to-audio generation AI platform called Stable Audio.

Stable Audio utilizes the diffusion model, an AI model used by the company's popular image platform Stable Diffusion, but this time trained on audio instead of images. Users can use the platform to generate songs or background audio for any project.

Audio diffusion models typically generate fixed-length audio, which is not ideal for music production as songs can have varying lengths. Therefore, Stability AI's new platform allows users to generate sound of different lengths, requiring the company to train the model on music and add text metadata at the beginning and end of the songs.

Previously, models trained on 30-second audio clips could only generate 30-second audio and randomly compose song fragments. Stability AI states that with adjustments made to the model, users of Stable Audio now have better control over the length of the songs.

"Stable Audio represents the cutting-edge audio generation research achievements of Stability AI's Harmonai Generative Audio Research Lab," the company stated in a press release. "We will continue to improve the model architecture, dataset, and training procedures to enhance output quality, controllability, inference speed, and output length."

According to the company, Stable Audio was developed by training on a dataset of 800,000 audio files containing "music, sound effects, and monophonic tracks," along with text metadata from music licensing company AudioSparx. The dataset amounts to over 19,500 hours of sound. Through collaboration with licensing companies, Stability AI claims to have obtained permission to use copyrighted materials.

Stable Audio will be offered in three pricing tiers: a free version allowing users to create up to 20 audio tracks of 45 seconds each per month, a professional version priced at $11.99 that enables the creation of up to 500 audio tracks of 90 seconds each, and an enterprise subscription that allows companies to customize according to their needs and pricing. Users of the free version are not permitted to use audio generated by Stable Audio for commercial purposes.

Text-to-audio generation is not a new concept, and other prominent companies in the generative AI field have been researching this concept. Meta released AudioCraft in August, a generative AI model suite that helps generate natural-sounding environmental sounds, voices, and music based on prompts. Currently, it is only available to researchers and some audio professionals. Google's MusicLM also allows users to generate audio but is only open to researchers.

Like other generative AI audio platforms, one potential use of Stable Audio will be to provide background music for podcasts or video production to enhance productivity.

Last year, Stability AI announced plans to expand its business into the audio generation, video, and 3D image domains.

RECENT AI NEWS

RECENT AI TOOLS