Position also matters! Separating Same Instruments in String Quartet using Timbral and Positional Cues

String quartets inherently face challenges such as fundamental frequency overlap, similar timbres, and complex acoustic environments, making existing source separation methods relying solely on timbral characteristics less effective. This paper proposes a novel approach that enhances separation performance by encoding timbral and spatial positional cues to guide a neural network for source separation. To evaluate our method, we created a dataset of string quartet recordings with realistic reverberation with two different seating arrangements. Experiments demonstrate that our approach outperforms current state-of-the-art (SOTA) methods in source separation, achieving superior performance even in seating arrangements different from training, thus highlighting the importance of our timbral and positional encoding in improving separation quality and generalizability.