I\'ve found an article that suggests:
def rms(sound, rms_level=-5): # sound.shape = (samples, channels) r = 10**(rms_level / 10.0) a = np.sqrt((le