# conversion of voice to frequency domain using fft

Started by October 21, 2008
```Hai to all,
Iam new to this forum and dsp. Iam involved in converting a 5sec voice
to frequency using a fft. Using pic24f series i need to accomplish this.
Can anyone guide through this process? What should be my sampling rate, How
can i convert the digital samples into real and imaginary parts, since the
c coded fft requires real and imaginary parts as arguments to fft.How many
point fft should i implement?

Siva

```
```sivaramakrishna wrote:
> Hai to all,
>   Iam new to this forum and dsp. Iam involved in converting a 5sec voice
> to frequency using a fft. Using pic24f series i need to accomplish this.
> Can anyone guide through this process? What should be my sampling rate, How
> can i convert the digital samples into real and imaginary parts, since the
> c coded fft requires real and imaginary parts as arguments to fft.How many
> point fft should i implement?
>
> Siva

Sorry for my bad english, i am here in germany

First, You need a low-pass filter. Then your sample rate has to be
faster than twice the maximum frequency which comes throug your filter.

For the calculation, you can simly take the imagimary values of the
input signal as zero. With FFT, you then get a complex spectrum. From
this, you can calculate the power-spectrum by x = const *(real*real +
imag*imag).

Stefan
```
```>sivaramakrishna wrote:
>> Hai to all,
>>   Iam new to this forum and dsp. Iam involved in converting a 5sec
voice
>> to frequency using a fft. Using pic24f series i need to accomplish
this.
>> Can anyone guide through this process? What should be my sampling rate,
How
>> can i convert the digital samples into real and imaginary parts, since
the
>> c coded fft requires real and imaginary parts as arguments to fft.How
many
>> point fft should i implement?
>>
>> Siva
>
>Sorry for my bad english, i am here in germany
>
>First, You need a low-pass filter. Then your sample rate has to be
>faster than twice the maximum frequency which comes throug your filter.
>
>For the calculation, you can simly take the imagimary values of the
>input signal as zero. With FFT, you then get a complex spectrum. From
>this, you can calculate the power-spectrum by x = const *(real*real +
>imag*imag).
>
>
>Stefan
>

My pic24f controller has a 22k program memory and 8k data memory.  I need
to implement a fft with this hardware.  So please find me a c code for fft
which makes the micro controllers performance optimum.

Thanks,
Siva
```
```"sivaramakrishna" <siva.gram@gmail.com> wrote in message
news:8_mdnTybNJKVSGPVnZ2dnUVZ_i2dnZ2d@giganews.com...
>
> My pic24f controller has a 22k program memory and 8k data memory.  I need
> to implement a fft with this hardware.  So please find me a c code for fft
> which makes the micro controllers performance optimum.

Meindert

```
```>"sivaramakrishna" <siva.gram@gmail.com> wrote in message
>news:8_mdnTybNJKVSGPVnZ2dnUVZ_i2dnZ2d@giganews.com...
>>
>> My pic24f controller has a 22k program memory and 8k data memory.  I
need
>> to implement a fft with this hardware.  So please find me a c code for
fft
>> which makes the micro controllers performance optimum.
>
>
>Meindert
>
>
>

But i could not find a one which suits my requirement.

Siva
```
```"sivaramakrishna" <siva.gram@gmail.com> wrote
> But i could not find a one which suits my requirement.

Look for any FFT code on the web like fftw (google+fftw)
Don't forget to put an antialiasing filter or you will get spectrum
overllaping!

For voice signal 3kHz filter should be fine, and you should sampe with at
least 7kHz ! Number of samples depend on you're wanted physical spectrum
resolution. For radix-2 algorithm try 64/128/256 and see what suits you
best.

Also, it would be great to learn basics of DSP :) Greetings from Croatia
from a student!

```