Interpolating music samples has the potential to enhance both the research workflow and the creativity of musicians. For the former, it can be used for augmentation, as data scarcity is a frequent hindrance to MIR research, whereas musicians can discover interesting soundscapes to incorporate in their performances and compositions. There are vast possibilities to perform the same musical phrases not only with different timbre but also with different tempo, timing, and accents. We approach this problem through the lens of the embedding space Music2Latent. Our preliminary work empirically explores how interpolated latents correspond to different musical properties. To ensure temporal consistency, we apply classical interpolation algorithms that are devised to handle sequences of different lengths.