Nope! A Rockchip RK3588 won’t be enough to perform complex AI computations for real-time stem separation.
I asked ChatGPT, and according to his explanation, an RK3588 has an AI computing power of around 10 TOPS. For high-quality real-time stem separation comparable to Virtual DJ Stems 2.0, you’d need an SoC with a computing capacity of at least 30 TOPS in a streaming pipeline. This means that the separation would be done segment by segment as the track plays. For a complete separation in less than 20 seconds with fast track loading, you’d need an SoC with a computing power of 45/50 TOPS, the equivalent of an M1 or a Snapdragon Elite.
The best and cheapest alternative in terms of production cost at this stage would be a Dimensity 9400+ (around 33 TOPS), which costs around $150 to purchase.
Add to this, you would need a total memory capacity of 12 GB of RAM (including 4 GB for running the operating system + 8 GB dedicated to AI calculations and decompression of stems in memory).
Rockchip currently does not have a sufficiently powerful SoC in its range.