--- a
+++ b/cnnmodel/feature_extraction/mfcc_extraction.py
@@ -0,0 +1,22 @@
+import numpy as np
+import librosa
+
+
+def get_mfcc(signal, samplerate):
+    # in librosa the window length and step size (stride) are set by number of frames and not
+    # duration. window_length is set by n_fft and step is set by hop_length
+    frame_length = int(0.025 * samplerate)
+    step_size = int(0.01 * samplerate)
+    mfcc = librosa.feature.mfcc(signal, samplerate, n_mfcc=13, n_fft=frame_length, hop_length=step_size, center=False)
+    mfcc_derivative = librosa.feature.delta(mfcc, order=1)
+    mfcc_second_derivative = librosa.feature.delta(mfcc, order=2)
+
+    assert mfcc.shape == (13, 10)
+    assert mfcc_derivative.shape == (13, 10)
+    assert mfcc_second_derivative.shape == (13, 10)
+
+    # stack mfcc, derivative and second derivative horizontally
+    mfcc_matrix = np.concatenate([mfcc, mfcc_derivative, mfcc_second_derivative], axis=1)
+    assert mfcc_matrix.shape == (13, 30)
+
+    return mfcc_matrix