FFT feature request

[jhidley]

It would be nice if a new visualization function could be added to the Control Panel. It would be average Spectrum. An FFT is processed over all of the data, or in the case of the second play button, all of the data that is selected (blue). The output is one frequency spectrum. To generate this, all of the individual FFT frequency bins are averaged, and the averaged value for each one is the new data point. Visually nothing would be visible until the audio file was played back, then the average spectrum would become visible.

This is extremely helpful when trying to look at just the signal and eliminate noise. The noise will have random phase, so it will cancel out, where the signal will be in phase with itself and add. This greatly increases the s/n ratio of the resulting analysis.

[DewDude420]

That holds true to a degree; but if your content is stereo, there is some phase randomization that occurs.

The only situation I can think of where you would have "stereo" noise would be FM MPX Stereo...where the noise from difference is out of phase in the final product. But the noise isn't the only thing out of phase...some of the stereo content is in there too.

Not to mention FFT is done per channel and isn't mixed in any way. An average display as a visualisation would have one average per channel.

[DougDbug]

It would be nice if a new visualization function could be added to the Control Panel. It would be average Spectrum. An FFT is processed over all of the data, or in the case of the second play button, all of the data that is selected (blue). The output is one frequency spectrum. To generate this, all of the individual FFT frequency bins are averaged, and the averaged value for each one is the new data point. Visually nothing would be visible until the audio file was played back, then the average spectrum would become visible.

Sorry, but I don't understand what you're saying. When you average you loose detail & information. That averaged information might be useful, depending on what kind of analysis you're doing. But once you've averaged (over a long period of time*) you no longer have anything like music or singing/speaking, etc., and reverse FFT will give you a mess.

This is extremely helpful when trying to look at just the signal and eliminate noise. The noise will have random phase, so it will cancel out, where the signal will be in phase with itself and add.

Phase needs a reference. We can talk about the phase between two signals of the same frequency, or we can talk about a phase-shift of a delayed signal relative to the original non-delayed signal. Or, if you invert the phase (polarity) of one stereo channel the sounds that are identical (or similar) and in-phase in both channels will tend to cancel (especially the bass) when the out-of-phase soundwaves mix in the air. But, if there is no similar sound in the opposite channel, phase has no meaning and nothing will happen when one channel is inverted.

One signal (or the noise with that signal) doesn't have any phase by itself.

This greatly increases the s/n ratio of the resulting analysis.

First, you have to define the noise (or the signal). Technically, noise can be something (usually random) that's added to the signal. Or, it may be something "unwanted" that's part of the signal. I have a recording of ocean wave sounds... That's "signal" but in an FFT analysis it's going to "look like" noise, and it actually sounds like noise, but since it's intentional it's not noise. The same thing with "T" and "S" sounds, or certain percussive sounds... They are very noise-like but they are only heard as noise if they come at the wrong time.


* Actually, averaging is part of the FFT process (the FFT sample size).

[jhidley]

Let me clarify what I'm looking for.

For this particular purpose I'm exciting a system with energy multiple times and measuring its response acoustically. The recording records the response of the device under test and any sounds in the environment which are noise.

When an FFT is performed on the recording, the excitation events will be in phase with each other and the noise will have random phase. This assumes that the environmental noise is really random and doesn't have a periodic nature to it (uncorrelated). The result of this is that when each FFT is summed, the signals add as they are in phase and the noise components add less (have some destructive interference) as the noise has random phase. For every time the device under test has its number of excitations doubled, the s/n ratio increases 3dB. This is a very common measurement technique used in acoustics. It is generally referred to as synchronous averaging.

I tried to post a link to more information about this technique, but the BBS won't allow it as I'm too new.

[DougDbug]

If you can handle the math you might try using MATLAB (or one of the free MATLAB clones). This is beyond anything I could do, but MATLAB can read WAV files and do FFT, and just about any other mathematical operations can imagine, and then graphically display the results if that's what you want.