AI Startup Stability AI has released Stable Audio Open Small. This is the “stereo” audio generation AI model that the company claims is the fastest on the market and efficient enough to run on smartphones.
Stable Audio Open Small is a collaboration fruit between chip manufacturers that produce stability AI and many processors within ARM, tablets, phones and other mobile devices. Many AI-powered apps can generate audio such as Suno and Udio, but most of the time they rely on cloud processing and cannot be used offline.
Stability also claims that the stable audio open small training set is made up entirely of free music archives and freesound songs from the royalty-free audio library. This includes copyrighted content, in contrast to the aforementioned Suno and Udio training set, which poses IP risk.
The Stable Audio Open Small has a size of 341 million parameters and is optimized to run on an ARM CPU. (A parameter, sometimes called weight, is the internal component of the model that guides its behavior.) Designed to quickly generate short audio samples and sound effects (such as drums and instrument riffs), the stable audio open small can generate up to 11 seconds of audio in under 8 seconds on a smartphone.
The samples generated by stable audio open small are as follows:
And here’s one more:
The model is not without limitations. Stable Audio Open Small only supports prompts written in English, and the documentation supports the document that the model cannot produce realistic vocals or high quality songs. The model also doesn’t work in the musical style as well, Stability warns. This is the result of biased Western training data.
Another potential wrinkle for developers, stable audio open small has a somewhat restrictive use term. It can be used for free to researchers, enthusiasts and businesses with annual revenues under $1 million, but developers and organizations that earn more than $1 million will have to pay for a stability enterprise license.
Stability, a plagued company behind the stable spread of popular image-generating models, raised new cash last year as investors, including Eric Schmidt and Napster founder Sean Parker, tried to turn the business around. Stability co-founder and EX-CEO Emad Mostaque has mismanaged stability to financial ruin, causing staff to step down, leading a partnership with Canva, and investors are reportedly concerned about the company’s outlook.
Over the past few months, Stability has hired a new CEO, appointed Titanic Director James Cameron to the board, and released several new image generation models.