Convolution Reverb help!

I'm developing a convolution reverb in Simulink by FFT'ing both input signals (Input from mic and recorded impulse response), then convolving them with each other in the frequency domain, and applying an IFFT before sending it to a speaker output. However, no matter how much i've tried, i only get noise out of it. I was wondering if someone could help me.

I've also tried to just convolve the signals in the time domain but i get the same result.

Any kind of thoughts or pointers would be much appreciated.

Thank you.

[José Goulart]

Felipe,

How are you computing the convolution? Please show us the code that you're using.

[Felipe Tavera]

Jose, right now I'm not using any code written by me on the model, just using the simulink blocks. I'm having one "from audio device" block connected to and fft block, then parallel to this I have the impulse response in a " from wave file" block connected to another fft block. These are joined ahead by a convolution block in the frequency domain. After this I'm connecting an ifft block, a block that has the real component of the signal for an output, and, finally the to audio device block.

I'm currently not at my usual workplace, but as soon as I get there I'll try to see if it's possible to upload a picture of the model or anything that proves to be more helpful than describing the blocks connection.

Thank you.

Answer 1

[Daniel Shub]

How much energy is there in the tail of your impulse response? If there is lots of late reverb, then the result is going to sound an awful lot like noise. What happens if you cut the duration of your impulse response (IR) down to only include about 50 ms of reverb?

[Felipe Tavera]

Daniel, right now i'm trying to do the process with an IR of 29.5ms of length @ a sample rate of 44100 (13021 samples). I figure it's length might not be the problem as I used this IR to calculate the reverberation and acoustic parameters in other software.

I'm really in a puzzle here as i don't know what might be wrong.

[Daniel Shub]

30 ms seems really short. Does your IR look qualitatively correct (1 big peak, a short gap and then a handful of other peaks). What happens if you make an artifical IR with just a peak and then 29.5 ms of zeros.

[Felipe Tavera]

Daniel, i'm sorry I didn't answer sooner, I'm seem to be having trouble with the email notifications from the matlab answers. The impulse response that I've been using looks qualitatively correct, is an initial peak with a short gap( ITGD) and the a reverberant tail. I'm going to try and do it with the artificial impulse response and see what happens from there, and report it back.

Thank you.

[Sean de Wolski]

Comments don't email you...