The thing is that the compression score looks lousy to me. For a signal with 32400 samples I obtain
If I am using fft for the compression, means: keeping only the highest coefficients of the Fourie transform of the signal and than take the inverse fft. I obtain L2 norm squared around 99 for the same compression ratio. My questions are: 1) Does the values for PERF0 and PERFL2 looks reasonable? 2) If so, is there some intuition of why compression with fft do much better? 3) Are there different ways to do audio compression with wavelet that can give better results?