0
$\begingroup$

I am making preparations to implement FFT on a wav File.wav Only Java. I'm reluctant to use matlab

I have a File wav stereo, 16 bits.

I need to calculate the FFT, but I have confusing.

My File has a Frequency of 44100, the samples in bytes are 10 minuts -> 

(10 minuts)*(60 seconds)*(44100 sample rate)*(2 bytes because are 16 bits)
= 52920000 >~ 52MB per channel.

As you can see it's a big file.

1. Question. Can I to split my file in order to make smaller calculus? If yes, My file can be variable, How I can calculate a moderate lenght working in function of frequency (greater frequency will need greater length FFT).

2. Question. According to FFT the size need to have a lenght = mantissa = Pow(2, Integer). Can I to fill with zeros to reach Pow(2, Integer), Example I have a file with a lenght of 65000 samples, then I will need a FFT with order of 16 -> mantissa = 65536, Can i to fill with zeros the others 536 samples?

Thank you.

Improve this question
$\endgroup$
4
  • 1
    $\begingroup$ Can you explain the goal of calculating of FFT for whole file? It is possible that you need STFT (Short time Fourier transform) - en.wikipedia.org/wiki/Short-time_Fourier_transform . About your questions. The frequency resolution depend on length of input array. Maximum frequency depend on time step. It is good idea to read textbook second time before implementing FFT. I think there are a lot java FFT library (see for example - stackoverflow.com/questions/3287518/…). Why do you need implement it? (homework?). $\endgroup$
    – SergV
    Commented Oct 11, 2013 at 4:20
  • $\begingroup$ Thank you.__________Can you explain the goal of calculating of FFT for whole file?------> I want to extract the main frequency of Sample, example of sound of piano recorded into file.__________ Why do you need implement it?------>I want to implement something like: WaveSpectra -> s000.tinyupload.com/… But I need Full control of process and my data. Now, I need implement STFT. $\endgroup$
    – Luigi Giuseppe
    Commented Oct 11, 2013 at 14:46
  • 1
    $\begingroup$ SergV: rework your comment into a proper answer, maybe by elaborating a bit more on why it doesn't make sense to compute the FFT of 10 minutes of audio (concept of stationarity), and maybe give a short overview of the STFT computation process (with references to windowing and zero-padding, maybe a warning about the time/frequency resolution trade-off), you'll get an upvote on that :) $\endgroup$
    – pichenettes
    Commented Oct 11, 2013 at 16:47
  • $\begingroup$ You can to obtasin WaveSpectra here: mediafire.com/download/uv92kugn52650wo/WaveSpectra.zip $\endgroup$
    – Luigi Giuseppe
    Commented May 19, 2014 at 3:15

1 Answer 1

Reset to default
2
$\begingroup$

Simply taking the Fourier transform of a ten minute signal is unlikely to provide you with any useful information - I suspect what is actually of interest to you is the spectral information at intervals throughout the signal, e.g. 600 frequency plots at 1 second intervals, which if you so desire can be combined into a spectographic image for analysis. This can be achieved by performing multiple windowed Short Time Fourier Transforms (STFT) and storing the results.

Can I to split my file in order to make smaller [calculations]?

Yes, it is up to you what granularity of frequency information you wish to obtain. Small FFT lengths will afford you a better time resolution, longer FFT lengths will afford you a better frequency resolution. For purely visualisation purposes an FFT length of 1024 is generally sufficient.

Can I to fill with zeros to reach Pow(2, Integer) [for my FFT length]?

Yes, if you do not have sufficient samples to reach a power of two length for your FFT then you should zero pad the input. This will introduce the minimal amount of spectral error into your frequency domain results.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thank you for your answer, Now It's clear that The FT isn't the best choice._________________ I was thinking about of granularity of representation of frequency it's a problem._________________ I had not heard of STFT neither Wavelet. Now I'm thinking in this last option._________________. Now my option will be Wavelet. _________________How can I refute (looking for more information or answer) in this site taking similar question avoiding the mark "duplicate question"? Thank you again for your time... $\endgroup$
    – Luigi Giuseppe
    Commented Oct 12, 2013 at 0:27
  • $\begingroup$ I think you can probably just re-ask the question provided the answer will be something different. I can't make any guarantees about it not being closed as a duplicate though... $\endgroup$
    – PAK-9
    Commented Oct 14, 2013 at 10:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.