A breakthrough in artificial intelligence technology has emerged as Stability AI unveils Stable Audio Open Small, an advanced audio-generating model. This innovative creation is touted as the fastest available and efficient enough to operate on smartphones. The result of a partnership with Arm, this model distinguishes itself from competitors like Suno and Udio by functioning without cloud reliance, allowing offline usage. Its training data originates solely from royalty-free sources, setting it apart from potential intellectual property risks associated with other models.
This stereo audio-generating marvel boasts 341 million parameters, specifically optimized for Arm CPUs. Designed for rapid production of short audio samples and sound effects, it can generate up to 11 seconds of audio within less than 8 seconds on a smartphone. Despite its impressive capabilities, the model has certain limitations. It only supports English prompts and struggles with realistic vocals or high-quality songs due to Western-biased training data.
The usage terms of Stable Audio Open Small present some restrictions. Free for researchers, hobbyists, and small businesses earning under $1 million annually, developers and larger organizations must acquire Stability’s enterprise license if their revenue exceeds this threshold. This stipulation adds a layer of complexity for potential users.
In recent developments, Stability AI, known for its popular image-generation model Stable Diffusion, secured new funding last year amidst leadership challenges. Following financial mismanagement by co-founder Emad Mostaque, the company hired a new CEO and appointed filmmaker James Cameron to its board. Alongside these changes, several new image-generation models have been released.
Stability AI's introduction of Stable Audio Open Small marks a significant advancement in AI technology. By enabling offline audio generation on mobile devices and utilizing exclusively royalty-free training data, the model addresses key concerns within the industry. Although limited in language support and vocal realism, it represents a leap forward in accessible AI tools. The licensing structure reflects an effort to balance accessibility with commercial viability, potentially reshaping how audio content is created and utilized.
