A Semantically Consistent Dataset for Data-Efficient Query-Based Universal Sound Separation - View it on GitHub