What can be the cause of getting so large values of components after ifft and deframing? ?

1)I'm working on speech frames.

Took 512 point fft of each frame.To process them further ,i took (N/2)+1 fft components of each frame making the very first component of each frame zero to reduce dc bias?

After processing the fft components separately as real and imaginary,i see that the real and imaginary parts are quite small.

But when i take ifft and do defaming the component values are quite high.It makes the time domain recovered speech signal too loud and also seg SNR too small. In output spectrogram frequencies above 1.5khz are not seen but also noise is not there,pesq good but very low segSNR.

What can be the cause of getting so large values of components after ifft and deframing? ?

[David Goodmanson]

Hi kalpana,

it sounds like you are taking the 257 nonnegative frequency components. What is the nature of the processing you are doing, and after that how do you get back to 512 points in the time domain?

Answer 1

[Vaibhav Tomar]

Hey Kalpana,

Wanted to ask a few things.

It would greatly help if you could provide some information regarding the processing that you are doing and the steps followed to get back to the 512 points in the time domain.

Thanks