Hello, when computing cepstrum why don't you compensate magnitude by norm1(window) applied i.e ifft(log(mag / norm1)) instead?