Differentiating the restitution of timbre by several loudspeakers may result from standard measurements, or from listening tests. This work proposes a protocol keeping a close relationship between the objective and perceptual evaluations: the stimuli are musical excerpts, and the measuring environment is a standard listening room. The protocol involves recordings made at a listener position, and objective dissimilarities are computed using an auditory model simulating masking effects. The resulting data correlate very well with listening tests using the same recordings, and show similar dependencies on the major parameters identified from the dissimilarity matrices. To cite this article: M. Lavandier et al., C. R. Mecanique 334 (2006).