Can someone tell me what is the starting premise in designing a FIR filter? Say, I want to filter human voice what should do? Do I pretend that the filter (be it butterworth, exponential, trapezoidal) that I am going to design is a filter to filter the human voice in frequency domain? In the Numerical Recipes, it states that one should pass the filter into a FFT to get the coefficients. For a FIR filter, one should try to use the minimum set of this FFT coefficients. The trick is to take the 1st K and the last K coefficients and zero out the rest. Then, move the last K to the front of the array and the 1st K should follow it, which leaves the remaining part of the array as zeros.
Take the FFT again on the newly reordered array and compare the absolute value of this FFT output with the original filter function. If this is deemed close then one will have a good filter.
Is this right so far? TIA. Please reply via email: firstname.lastname@example.org, email@example.com