Possibly. The reference to the frequency axis let me to think the way I explained, but without it I agree that Gilles' answer is more accurate.

Mehdi : yes. If you do a FFT in the conditions you are explaining, you will not be able to reconstruct the 25 Hz signal if the initial data had a 30 Hz sampling rate. @Jojek : That is what I understood from Mehdi's question : "Is this sampling rate the same as the frequency used when calculating the FFT when creating the frequency axis of the FFT?" means for me "Is this sampling rate the same as the frequency axis of the FFT". From Mehdi's answer above it seems to me that it was the kind of information he was looking for. If not, Mehdi maybe you can clarify the question ?

Oh okay - I had not realized those were coming from an IDE. I'll try to find something similar or will code it myself one day. Thank you.

Awesome, thank you very much. Can't believe I spent several days (with breaks of course!) with such an obvious mistake waiting to be solved. Very helpful, thanks a lot. By theway, which software are you using to display those nice graphs ?

Thanks for this. I've used the normalized coefficients with the same results, except that now there is not the volume change I mentioned previously. However I see/hear no filtering when feeding the plugin I'm building with white noise.

You're right, I edited my initial post. Whatever the coefficients are though, I believe my problem is mostly linked to the algorithm rather than the coeffs.