Piano body deconvolution
This is a simple experiment with blind piano body deconvolution using log-spectrum averaging of the whole sound FFT (>37s) for all sounds in a piano database.
- Here is the minimum phase IR extracted from all the highest velocities piano samples: piano body V5
- The minimum phase IR extracted from all the non sustained piano samples with highest velocity: piano body nosus v5
- The minimum phase IR extracted from all the release piano samples: piano body Rls
- The minimum phase IR extracted from all the C3 samples: piano body C3
- And the C3-v4 excitation deconvolved by piano_body_C3: German v4 C3 deconv
The extracted sounds obviously inherit the average EQ of the given sample set. But the results are not so bad given the simple means used.
References:
- A. Openheim, R.W Schafer, T. G. Stockham “Nonlinear Filtering of Multiplied and Convoled Signals”
- Mohamed Tria, Mirko Van Der Baan, Anthony Larue, Jerome Mars, “Wavelet estimation in homomorphic domain by spectral averaging, for deconvolution of seismic data”