STEMS is our family of stem generation algorithms, capable of separating any mono or stereo mix into six discrete stems: Vocals, Bass, Drums, Guitar, Piano, and Other. Additionally, enhanced neural separators can be applied to further refine the resulting stems.
This is optional for each stem, and allows users to tailor the balance between performance and audio quality to their own needs and preferences. It also allows for a reduced latency of 2800ms, as opposed to 12000ms. Even on slower devices, STEMS can be configured to deliver the appropriate performance.
STEMS PRO is the advanced version of the algorithm, which balances audio quality and performance to run on desktop applications. In standard mode, it runs at approximately 12 times real-time on an Apple Silicon M1 chip using only the CPU, and maintains the highest industry standards in real-time stems processing.
STEMS EFFICIENT is optimized for use on low-performance CPUs, and is much faster than other available stem separation algorithms while offering competitive quality. It runs at approximately 34 times real-time, depending on the device’s performance.