Developing an Effective Training Dataset to Enhance the Performance of AI-based Speaker Separation Systems