home *** CD-ROM | disk | FTP | other *** search
-
-
- What is an FFT?
-
- An Article
-
- by Nicholas Blachford
-
-
-
- You may of heard of an 'FFT'? It usually pops up in audio programs with
- about one line in the manual which says something like - "this gives an
- FFT" which is not the most useful of explanations! Unless you're into
- Electronics in a big way you probably won't know what they are so I'm
- going to spill the proverbial beans and tell you what they are...
-
- 'FFT' stands for Fast Fourier Transform, presumably these are normally
- slow so I suppose the fast means someone wrote a faster one. Fourier
- was a French mathematician who thought up a mathematical process, which he
- called a Fourier Transform, a mathematical method of breaking up a
- waveform into its various frequency bands. The theory goes, that every
- sound is made up of a series of different sine waves at different pitches
- (or Harmonics), the FFT is the method used to cut the sound into its
- various Harmonics.
-
- The result can be plotted in a graph with the frequencies along the X
- (bottom) axis and amount along the Y (sideways) axis. If you could get
- a synthesizer which could generate all these sine waves at there
- relevant frequencies and volumes you would then be able to recreate the
- wave you started with. The plot produced in the PD program FMSynth is like
- this.
-
- Sometimes you see FFTs plotted in a 3D form. As you may know, sounds
- change over time, and the 3D FFT is really a FFT plotted over time, with
- time on the Z axis (going into the screen). If you had a synth capable
- of following this graph you would have a machine capable of Resynthesis.
- Oberhiem are working on this for release next year but if you're
- impatient an existing system is already on sale for around รบ30.000!
-
- Using Resynthesis it would be possible to convert a sample into a very
- small file which could then be recreated at any time - a 16 bit sample at
- 48KHz takes nearly 100K of data, whereas a Resynthesis file would take a
- couple of K of memory so it is very efficient space wise; this is roughly
- equivalent to a CD with 65 hours (yes Hours!) of music or a standard Amiga
- floppy storing 3.6 minutes of stereo CD Audio. We could use this technique
- on the Amiga but it would take forever given the heavy computations
- involved - it may be possible given some of the new generation of RISC
- chips but special hardware would be required for anything more than a
- couple of channels.
-
-
- I've included an example of a 3D FFT picture along with the sample it
- came from. At the top of the picture is a 3D FFT plot. The Sound is
- called 'Sine-Square.iff' because it starts with a SINE wave which morphs
- into a SQUARE wave. Bass is on the left and treble on the right. The
- sound starts with some bass and no higher harmonics but these soon appear
- and then by the end the graph has become a row of spikes. The reason for
- this pattern, is that a sine wave has no additional harmonics while the
- square wave is full of them. Any digital sound output is made of square
- steps and it is the higher harmonics of these steps which make all the
- additional noise when a low sampling rate is used.
-
- So what do you do with them? I sometimes look at them to see what's going
- on in a sample and you can use them to get cleaner loop points by checking
- to see if the beginning and end of the loop match. Apart from that, FFTs
- are useful for analyzing sounds and looking good.
-
-
- Nicholas Blachford
-
- 31st December 1994
-