The acoustic and auditory properties of individual musical instruments have been extensively studied over recent decades, however, their sonic interplay within a musical ensemble remains under-explored. Given their considerable importance across various fields, aspects of ensemble sound, such as blending of instruments, perceptual relevance of directivity of instruments, and the role of room acoustics, demand comprehensive evaluations and an interdisciplinary approach. This study aims to improve the perceptually motivated acoustic representation of instruments in joint performance in both real and virtual acoustic domains, by exploring different stages of these aspects in musically realistic contexts. An explorative listening test with live string ensemble performance suggested that the characteristics of the acoustic environment considerably influence the blending of violins playing in unison. Combining methods of Machine Learning and Music Information Retrieval, a computational modelling approach is proposed to classify sound samples from an ensemble recording according to perceived blending. Proving this classification to be effective for monophonically rendered sound samples of two violins from in-situ environments, without requiring the individual source recordings marks a first step towards comprehensive blending modelling. Furthermore, the applicability of close-microphone recordings for auralization of a perceptually convincing ensemble sound was successfully demonstrated. Advancing previous research in directivity perception, it could be demonstrated that the room acoustics have a greater impact on the orientation perception of sources than their directivity. By involving instruments with distinct radiation directivities in a variety of acoustic environments, the major acoustical parameters influencing the orientation perception have been explored. Examining musical instruments with their inherent dynamic directivity against loudspeakers in in-situ conditions showed that their distinction becomes obscured under specific acoustic conditions. These findings led to a pilot study on the perceptual relevance of high-order directivity modelling of individual sources forming an ensemble. Results indicate, that even with an increasing number of sources, their detailed directivity characteristics remain pivotal for auralizing ensemble performance. The role of room acoustics in shaping the overall blending is shown to be dependent on the source-level blending. A computational model for predicting overall perceived blending in musical performance using source-level blending ratings and room acoustical parameters was suggested and validated. Analysis of its feature importance revealed that the room acoustic contribution to the overall blending impression is nearly as significant as the blending between instruments at the source level. By emphasizing and detailing relations between musical blending, directivity perception, and auralization aspects, this thesis contributes to the advancement of ensemble sound research and offers insights pertinent to music performance and perception research, virtual acoustics, and related fields.