Frequency analysis are different. (python numpy fft and frequency analysis of Audition)

[高山0D4A]

Frequency analysis was performed with the following Python code. 

def GetFreq(wav_file_path, start_sec, end_sec, N):
    overlap_pt = N//2       #50% overlap

    #read wav file
    ir, fs = sf.read(wav_file_path)
    st_sample = int(start_sec*fs)
    ed_sample = int(end_sec*fs)

    frame_num = (ed_sample - st_sample) // overlap_pt - 1

    #window
    window = np.blackman(N)
    acf = 1/(sum(window)/N)

    FFT_result_dB = np.zeros((0, N//2+1))
    for i in range(frame_num):
        FFT_result_dB = np.append(FFT_result_dB, np.zeros((1, N//2+1)), axis=0)
        
        st = st_sample+i*overlap_pt
        ed = st+N

        #windowed
        x_windowed = ir[st:ed]*window

        #FFT
        F = np.fft.fft(x_windowed)
        F = F * (2.0 / N)        #Normalization
        F = F * acf
        F_abs_amp = np.abs(F)
        F_abs_amp_dB = 20.0 * np.log10(F_abs_amp + 1.0e-10)
        FFT_result_dB[i] = F_abs_amp_dB[:N//2+1]

    #Averaging
    fft_out = np.average(FFT_result_dB, axis=0)

    freq = np.arange(0, fs, fs/N)
    freq_out = freq[:N//2+1]

    return [freq_out, fft_out]

if __name__ == "__main__":
    freq, fft1 = GetFreq('sin100Hz.wav', 1, 4, 512)
    freq, fft2 = GetFreq('WN.wav', 1, 4, 512)

Frequency analysis was performed in Audition at the same FFT point.

Compare the results of the two frequency analyzes.

For a 100Hz sine wave, the values are about the same. However, white noise has a different amplitude. 

For 100Hz sine wave

For white noise

Why does it make a difference?
Is there any technical document of frequency analysis by Audition?

Best regards.

[SteveG_AudioMasters_ · Accepted solution]

There are no technical details of any analysis that Audition does - never have been, probably never will be. All you have there by the looks of it is a 3dB difference in the noise floor. Without knowing exactly how you fed your signal into whatever you did the Python analysis with, I couldn't even hazard a guess as to why this is. And even then it will be a bit of a mystery, because the peak amplitudes look the same.

But it's no good asking - nobody's going to tell you anything about the analysis methods in Audition, because none of us know. And the developers (who won't see this thread anyway) wouldn't tell you because they regard all information like this as proprietary.