perceptual_weighting: loudness compensation for time-frequency representation. by Keunwoo Choi ([email protected])
- loudness.py: You can get perceptually weighted time-frequency representation in a much more correct way than A-weighting.
- basic_weightings.py: You can add A-, B-, C-, and D-weighting to your TF-representation.
- Because energy of tf-bin does not represent what we hear in terms of loudness.
You see the results in the included png files.
Please feel free to suggest me further upgrade!
As in the example.py.
- preparation
# some constants
sr = 44100
n_fft = 1024
bin_width = float(sr)/n_fft
freqs = [0.0] + [(i+1)*bin_width for i in xrange(0,n_fft/2)] # important!
freqs = np.array(freqs)
S = np.load('data/spectrogram.npy')- conversion
import loudness
converter = loudness.ISO226_Converter(freqs, isStrict=False) # initiate a converter object for the frequency bands
S_weighted = converter.convert_s2l(S) # convert the spectrogram.