Open MultiTrickFox opened 1 year ago
given the basic conversion: cqt = CQT_nsgt(numocts, binsoct, mode="matrix_complete", fs=fs, audiolen=Ls, device="cpu", dtype=torch.float32) coeffs = cqt.fwd(data) data = cqt.bwd(coeffs)
checking the coeffs array seems its not normalized; print(coeffs.real.min(), coeffs.real.max())
Is there a way to quickly and correctly fix this, including being used in bwd() later on too..
given the basic conversion: cqt = CQT_nsgt(numocts, binsoct, mode="matrix_complete", fs=fs, audiolen=Ls, device="cpu", dtype=torch.float32) coeffs = cqt.fwd(data) data = cqt.bwd(coeffs)
checking the coeffs array seems its not normalized; print(coeffs.real.min(), coeffs.real.max())
tensor(-101.7243) tensor(140.4326)
Is there a way to quickly and correctly fix this, including being used in bwd() later on too..