def extract_mel_spectrogram(wav_path, X, y, index, curr_speaker_num):
"""
Extracts the mel spectrogram into the X array and saves the speaker into y.
:param wav_path: the path to the wav file
:param X: return Array that saves the mel spectrogram
:param y: return Array that saves the speaker numbers
:param index: the index in X and y this is stored in
:param curr_speaker_num: the speaker number of the current speaker
:return: a one (1) to increase the index
"""
Sxx = spectrogram_converter.mel_spectrogram(wav_path)
for i in range(Sxx.shape[0]):
for j in range(Sxx.shape[1]):
X[index, 0, i, j] = Sxx[i, j]
y[index] = curr_speaker_num
return 1
def extract_mel_spectrogram(wav_path, X, y, index, speaker_uid):
"""
Extracts the mel spectrogram into the X array and saves the speaker into y.
:param wav_path: the path to the wav file
:param X: return Array that saves the mel spectrogram
:param y: return Array that saves the speaker numbers
:param index: the index in X and y this is stored in
:param speaker_uid: the speaker number of the current speaker (integer hash of his identifier)
:return: a one (1) to increase the index
"""
#print('processing ', wav_path)
Sxx = spectrogram_converter.mel_spectrogram(wav_path)
for i in range(Sxx.shape[0]):
for j in range(Sxx.shape[1]):
# In case the file is longer than the :max_audio_length defined in the speaker_factory.py,
# we only use the spectrogram up to that cut off point at :max_audio_length
if j >= X.shape[3]:
continue
X[index, 0, i, j] = Sxx[i, j]
y[index] = speaker_uid
def save_spectrogramm_png(path):
# Load the mel spectrogram
spectrogram = mel_spectrogram(path)
# Begin the plot
figure = plot.figure(1)
plot.imshow(spectrogram[:, 20:160])
# Add the color bar
color_bar = plot.colorbar()
n = np.linspace(0, 35, num=11)
labels = []
for l in n:
labels.append(str(l) + ' dB')
color_bar.ax.set_yticklabels(labels)
# Add x and y labels
plot.xlabel('Spektra (in Zeit)')
plot.ylabel('Frequenz-Datenpunkte')
# Save the figure to disc
figure.savefig(get_result_png('spectrogram'))