DSPRelated.com
Forums

How to deal with the overlap when constructing signal from frequency domain?

Started by hyee...@yahoo.com.cn December 2, 2008
All,here is 2 questions about constrcuting signal from frequency domain. Any comments are appreciated.

1. I transmit some speech signal,with no window, into frequency by fft, and do nothing and back to time domain by ifft at once. It sound good as the original siganl. But repeated above with hamming/hanning windowed signal,the constructed signal would induce some noise(Hiss or dithering). why?

some script here:
N %6;
M = N/2;
[x,fs] = wavread('origsign.wav');

s = buffer(x,N);
win = hanning(N);
win = win*ones(1,size(s,2));

%s = s.*win;
S = fft(s);
p = ifft(S);
y = p(:);
wavwrite(y,fs,'syn.wav');
s = s.*win;
S = fft(s);
p = ifft(S);
y = p(:);
wavwrite(y,fs,'syn1.wav');

2. Another question is about dealing with overlap when constrcuting signal from frequency.

Enframe speech signal with overlap, and transmit it into frquency ... ,and back to time. When constrcuting it,how to deal with the overlap parts? Just use one of overlap (overlap save),or add all of overlap parts(overlap add,it maybe 2,3...and more). ? If overlap add, should I scale it? how to scale? whether it should use overlap save or overlap add,it is related to signal windowed?

How to deal with the overlap when constructing signal from frequency domain?

cheers
hyeewang