I think recording the audio from the dreams would be a first and easier step. Ignoring L/R audio, recording audio would be a linear line of data. Recording visual would be a 2D line of data, which is lot more memory.