The modulated lapped transform (MLT) can be regarded as the fundamental transform in the field of audio coding. The MLT is, however, an orthogonal tranform, which poses the limitation of the synthesis transform matrix being simply the transpose of the analysis transform matrix. This requires the analysis and synthesis window functions to be the same. It is possible to generalize the MLT into the biorthogonal modulated lapped transform (BMLT), and gain more flexibility in the choice of the analysis and synthesis window functions. It is further possible to adapt the BMLT window functions for example to tune the energy compaction of the analysis transform. This paper addresses the requirements of perfect reconstruction when the BMLT transform windows are varied in time.