Automatic Gain Control (AGC) for audio signals in python, based on Dan Ellis' Matlab code.

The code is based in the original Matlab implementation in the above link. It is almost exactly the same, with the exception of the STFT and ISTFT functions, which I implemented from scratch. A sample WAV file (obtained from the original Matlab source code) is provided for testing.

The code depends on NumPy / SciPy.

import scipy.io.wavfile
import numpy as np
from agc import tf_agc

# read audiofile
sr, d = scipy.io.wavfile.read('speech.wav')

# convert from int16 to float (-1,1) range
convert_16_bit = float(2 ** 15)
d = d / (convert_16_bit + 1.0)

# apply AGC
(y, D, E) = tf_agc(d, sr)

# convert back to int16 to save
y = np.int16(y / np.max(np.abs(y)) * convert_16_bit)
scipy.io.wavfile.write('speech_agc.wav', sr, y)

D. Ellis (2010), "Time-frequency automatic gain control", web resource, available: http://labrosa.ee.columbia.edu/matlab/tf_agc/