Technology
Sony CSL develops Diff-A-Riff, an AI-powered music generation tool (a "stem generator") designed to assist musicians in creating instrumental accompaniments that naturally complement existing musical material. The system utilizes a Latent Diffusion Model (LDM) to generate high-quality audio tracks at a 48kHz pseudo-stereo resolution.
Artists using Diff-A-Riff can influence the generated accompaniments through practical input controls: either by providing reference audio tracks to indicate musical style, instrumentation, or mood, or by describing desired musical characteristics using textual prompts.
Useful links
- Demonstration videos: Main Demo Video, Rock 'n' Roll Demo, Urban Hip-Hop Demo, Disco Demo
- Companion website featuring additional audio examples and system details
- Technical details available in the research paper
- Further Prototypes are listed and detailed on the Sony CSL Music Team website. Artists are welcome to consider any of these additional tools for their project
Requirements for artistic team
members' skills
members' skills
Applicants must have a strong interest in:
- knowledge in music production
- basic knowledge of generative AI
A background in the following areas would be a plus:
- a technical framework (like max/python/cpp/others)