to debug kapre's spectrogram. copied from ipython notebook.

kapre_debug.py

import matplotlib
%matplotlib inline
import matplotlib.pyplot as plt
plt.style.use('ggplot')
import numpy as np
import librosa
import keras
from keras import backend as K
print(keras.__version__)
print('Keras backend: ', keras.backend._BACKEND, ' and image dim ordering: ', keras.backend.image_dim_ordering())
import kapre
if keras.backend._BACKEND == 'tensorflow':
    import tensorflow
    print(tensorflow.__version__)
else:
    import theano
    print(theano.__version__)
from keras.models import Sequential
from librosa import display

duration = 1.0
n_dft = 1024
border_mode = 'same'
n_hop = n_dft / 3
n_filter = n_dft / 2 + 1
power = 1.0
dim_ordering = 'tf'
if dim_ordering == 'th':
    ch_axis_idx = 1
else:
    ch_axis_idx = 3
sr = 32000

# Load whatever
src, sr = librosa.load('../srcs/bensound-cute.mp3', mono=False, duration=duration, sr=sr)
src.shape

from kapre.time_frequency import Spectrogram
model = Sequential()
model.add(Spectrogram(n_dft=n_dft, n_hop=n_hop, input_shape=src.shape, power_spectrogram=2.0))
stft_per_channels = model.predict(src[np.newaxis, :, :])
# stft_per_channels = librosa.logamplitude(stft_per_channels, ref_power=1.0)
print('Shape of a batch (batch_size == 1 though): {}'.format(stft_per_channels.shape))

import scipy
window = scipy.signal.hann(n_dft, sym=False)
stft_librosa_l = np.abs(librosa.stft(src[0], n_dft, n_hop, window=window))**2
                                      
stft_librosa_r = librosa.logamplitude(np.abs(librosa.stft(src[1], n_dft, n_hop, window=window))**2, 
                                      ref_power=1.0)
print("shape of librosa result: {}".format(stft_librosa_l.shape))



librosa.display.specshow(np.abs(stft_per_channels[0, :, :, 0] - stft_librosa_l)[10], y_axis='linear', sr=sr)
plt.title('residual')
plt.colorbar()

In this line:
librosa.display.specshow(np.abs(stft_per_channels[0, :, :, 0] - stft_librosa_l)[10], y_axis='linear', sr=sr) there is a 10. I think this should be removed, isnt it?
Or maybe you wanted to write [:10] in order to put in evidence the differences?

Anyway, I got the idea. So thanks for the gist.