Real-Time Convolution with FFT within a Plug-In

16 posts

RELATED
PRODUCTS

Real-Time Convolution with FFT within a Plug-In 2012-05-24T18:18:20+00:00 Hey folks! I try to program a little audio unit plug-in that convolves the audio track with an impulse response given by the plug-in. i know that i need an fft algorithm that transfers the audio file's real signals to their spectra,multiplies the two spectra of the audio files and then returns them to their real signals. studying the "Audio Programming Book" i found an fft-based convolution algorithm that covers this operation. unfortunately this algorithm does not cover REAL-TIME convolution, which is of course absolutely necessary for a plug-in (is it?). nevertheless the book says: "This function will not serve for real-time purposes. However, it can be modified for this kind of application, by processing one FFT block at a time and leaving the overlapp-add to be performed by the invoking code ." (i'll post the code at the end of my post!) alright..concerning this i have a couple of questions that i just can't figure out myself, since i'm pretty new to this programming-audio-units-thing. 1. what exactly do i need to change within the code in order to make this function work within a real-time plug-in? maybe somebody can rewrite it or add the stuff needed if it isn't much work, because now i'm not making any progress. if not, maybe someone can show me the place in the code that needs to be changed and tell me how it needs to be changed. 2. how and at which location within the audio unit code do i load the impulse response? i couldn't figure that out, because i didn't find any examples that showed how to load a sound and where to load it within the code. 3. is there someone that knows audio units well or at least has some basic knowledge of them that can answer me further questions concerning that topic? that would help me A LOT! i read all i could find about them but some little things still aren't clear to me and i don't want to post a list of 20 lil questions in here. (maybe via email or personal message? would be just AWESOME) since this is concerning my bachelor thesis, any help would be marvellous :) tl;dr anyone familiar with real-time convolution and/or audio units? holla at me :) thank you guys very much in advance and greetings from germany! patrick void convol(float* impulse, float* input, float* output, int impulse_size, int input_size){ float *impspec, *inspec, *outspec; // spectral vectors float *insig, *outsig, *overlap; // time-domain vectors int fftsize=1, convsize; // transform and convolution sizes int overlap_size; // overlap size int count, i, j; // counter and loop variables overlap_size= impulse_size - 1; convsize = impulse_size + overlap_size; while(fftsize < convsize) fftsize *= 2; impspec = new float[fftsize]; // allocate memory for inspec = new float[fftsize]; // spectral vectors outspec = new float[fftsize]; insig = new float[fftsize]; outsig = new float[fftsize]; overlap = new float[overlap_size]; // get the impulse into the FFT input vector // pad with zeros for(i = 0; i < fftsize; i++){ if(i < impulse_size) insig[i] = impulse[i]; else insig[i] = 0.f; } // Take the DFT of impulse fft(insig, impspec, fftsize); // processing loop for(i = count = 0; i < input_size+convsize; i++, count++){ // if an input block is ready if(count == impulse_size && i < (input_size+impulse_size)){ // copy overlapping block for(j = 0; j < overlap_size ; j++) overlap[j] = outsig[j+impulse_size]; // pad input signal with zeros for(j = impulse_size; j < fftsize; j++) insig[j] = 0.f; // Take the DFT of input signal block fft(insig, inspec, fftsize); // complex multiplication // first pair is re[0Hz] and re[Nyquist] outspec[0] = inspec[0]*impspec[0]; outspec[1] = inspec[1]*impspec[1]; // (a+ib)*(c+id) = (ac - bd) + (ad + bc)i for(j = 2; j < fftsize; j+=2){ outspec[j] = inspec[j]*impspec[j] - inspec[j+1]*impspec[j+1]; outspec[j+1] = inspec[j]*impspec[j+1] + inspec[j+1]*impspec[j]; } // IDFT of the spectral product ifft(outspec, outsig, fftsize); // zero the sample counter count = 0; } // get the input signal // stop when the input is finished if(i < input_size) insig[count] = input[i]; else insig[count] = 0.f; // overlap-add output starts only // after the first convolution operation if(i >= impulse_size) output[i-impulse_size] = outsig[count] + (count < overlap_size ? overlap[count] : 0.f); } // de-allocate memory delete[] overlap; delete[] outsig; delete[] insig; delete[] outspec; delete[] inspec; delete[] impspec; } [/code] bountyhunter2010 https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=279730

xoxos xoxos https://www.kvraudio.com/forum/memberlist.php?mode=viewprofile&u=2639: Banned; 12367 posts since 30 Apr, 2002 from i might peeramid

Post by xoxos » Sat Dec 29, 2012 5:49 pm

0
Quote

note: thread raised from the dead

you come and go, you come and go. amitabha neither a follower nor a leader be tagore "where roads are made i lose my way" where there is certainty, consideration is absent.

Post Reply

16 posts

Return to “DSP and Plugin Development”