Audio equalization

Question

I'm developing audio player using FFmpeg and I want to add audio equaliqer to my app. I use FFmpeg to get audio samples and compute FFT, but when I try to apply one of IIR filters, I'm geting very noisy audio signal. This is my code:

double Q = 1.0;
double omega = 2.0 * PI * 1000.0 / 44100.0;
double sine = sin(omega);
double alpha = sine / ( 2.0 * Q);
double cosine = cos(omega);

double b0 = (1 + cosine)/2;
double b1 = (-1) * (1 + cosine);
double b2 = (1 + cosine)/2;
double a0 = 1 + alpha;
double a1 = (-2) * cosine;
double a2 = 1 - alpha;

for( int n = 2; n < fftSize; n++ )
{
   leftChannel2[n].re = ((b0/a0)*leftChannel[n].re + (b1/a0)*leftChannel[n-1].re + (b2/a0)*leftChannel[n-2].re -
                         (a1/a0)*leftChannel2[n-1].re - (a2/a0)*leftChannel2[n-2].re);
   rightChannel2[n].re = ((b0/a0)*rightChannel[n].re + (b1/a0)*rightChannel[n-1].re + (b2/a0)*rightChannel[n-2].re -
                          (a1/a0)*rightChannel2[n-1].re - (a2/a0)*rightChannel2[n-2].re);                    
   leftChannel2[n].im = leftChannel[n].im;
   rightChannel2[n].im = rightChannel[n].im;
}

Can anybody told me what is wrong with this code?

I really don't understand what are you trying to do using this code. Do you want to make a certain frequency range quieter or louder? — Andrey Nasonov, Oct 02 '15 at 13:07
@Andrey, According to Audio-EQ-Cookbook, this is LowPass filter with 1kHz cutting frequency. And I've missed some code. — Evgeniy Kharchenko, Oct 02 '15 at 13:27
Maybe I'm missing something, but shouldn't `sine` and `cosine` be functions, not one-off definitions? — underscore_d, Oct 02 '15 at 13:33
Actually according to [this Audio-EQ-Cookbook](http://www.musicdsp.org/files/Audio-EQ-Cookbook.txt), it is an HighPass filter (which is also what you get when plotting the filter frequency response with eg. `freqz` in Matlab). — SleuthEye, Oct 02 '15 at 13:34
@underscore-d, **sine** and **cosine** are variables to store sine and cosine of **omega**. — Evgeniy Kharchenko, Oct 02 '15 at 13:38
I wonder then why do you apply this filter to FFT coefficients. — Andrey Nasonov, Oct 02 '15 at 14:42
Just look here: https://en.wikipedia.org/wiki/Digital_biquad_filter how to apply it — Andrey Nasonov, Oct 02 '15 at 14:43
@Andrey, So I have to apply filter to audio samples directly? — Evgeniy Kharchenko, Oct 02 '15 at 15:13

score 0 · Accepted Answer · answered Oct 02 '15 at 14:03

0

Does this filter perform correctly in Excel or Matlab? At first look, i don't see here syntax or semantic errors. By the way, this filter (difference equation) computes frequency response in time domain. What about the imaginary part of the signal? If it is non-zero, you have to filter it in the same way.

answered Oct 02 '15 at 14:03

romanetz

16

Are you sure about time domain? I'm filtering a result of Fast Fourier Transform, which is in a frequency domain. And I tried to filter imaginary part of the signal in the same way, but nothing changed. – Evgeniy Kharchenko Oct 02 '15 at 14:31
You should apply this filter to time domain directly. Biquad filter is a recursive filter that cheaply approximates expensive filtering in a frequency domain. – Andrey Nasonov Oct 02 '15 at 14:45

Audio equalization

1 Answers1