Subject:Meaningful SpecAnalysis results

Posted by: Geoff_Wood

Date:2/17/2004 10:31:24 PM

I have a project which includes re-recording pink noise through a variety of mics, then getting an 'averaged' SA plot (say over a 5 second period). Anything I have tried only ends up with an instantaneous 'picture' of whent he sample stops.

Any ideas as to how to get an averaged plot over a time period ?

geoff

Subject:RE: Meaningful SpecAnalysis results

Reply by: Rednroll

Date:2/18/2004 8:00:47 AM

Yes, I do this all the time. You'll need a program like Spectra or Smaartlive. The Sound Forge SA is lacking in that department, which means it's virtually useless.

Subject:RE: Meaningful SpecAnalysis results

Reply by: Sonic

Date:2/18/2004 9:55:57 AM

Unless one of the realtime monitoring options is enabled, Spectrum Analysis will do an overlapped "average" analysis on any audio selection in the current file's window. Overlap and length parameters are accessible in SA settings.

So turn off Real Time Monitoring, record the mic input into Sound Forge, close the record dialog, select a duration of the recording you are interested in, then click SA refresh if auto-refresh is not enabled.

Use snapshots to store your results and you can compare multiple analyses. Or if it makes sense, you can mix multiple recordings and analyze the composite.

Or am I misunderstanding you?

J.

Subject:RE: Meaningful SpecAnalysis results

Reply by: Rednroll

Date:2/18/2004 2:58:21 PM

J.
I believe what you are describing with the overlap feature is an averaging of the X-axis(frequency) of the spectral analysis. What you need is an averaging of the Y-axis (amplitude), so you get more of an average curve of the frequency "response", instead of instantaneous snap shot. Unless, I'm misunderstanding the overlap feature, this doesn't seem to give what is needed.

Red

Subject:RE: Meaningful SpecAnalysis results

Reply by: Sonic

Date:2/19/2004 8:03:05 AM

The overlap feature is standard overlap-add short-term fourier analysis. If you make a selection greater than the length of the FFT, the N-point FFT is taken every (N - overlap) samples within that region and accumulated with standard windowing techniques. So the display *is* the conglomerate ("average") frequency response of all data in the selection, not any one single N-point FFT.

Not sure what you mean by axes. The x-axis range is predetermined by the sample rate and the bin width is predetermined by the FFT size.

Perhaps you would rather see how the response changes over time. In that case, you might try the sonogram mode. Alternatively, you can increase the number of 'slices' in SA settings, which will take FFTs at regular intervals and display them in 3d fashion.

J.

Subject:RE: Meaningful SpecAnalysis results

Reply by: Rednroll

Date:2/19/2004 8:44:53 AM

What you described makes sense to me, I will play around with the overlap setting and see if I can get what I'm looking for.

"Perhaps you would rather see how the response changes over time. In that case, you might try the sonogram mode."

No not at all. Maybe my X/Y description was inaccurate. I've tried to use this analogy in the past, I'll give it another shot. What where looking for is an averaging setting, much the way a VU meter shows an average loudness. For frequency response, I should be able to slow the averaging much slower than a VU meter does. With an averaging setting, I should be able to record pink noise of an acoustical environment. Then when I do the spectral analysis, the frequency curve should be nearly identical if I'm taking a snapshot at 1 second or 5 seconds later. Because eccentially I've captured the average acoustical response. Currently I can take different snapshots at different times and see way too much of a variance between snapshots to give me an idea of how the acoustical environment is actually effecting audio frequencies. Too me this acts much like a peak meter works, in that the displayed value is instantaneous. Which is good if that's what I'm trying to get, but I want a snapshot of the average frequency response.

Subject:RE: Meaningful SpecAnalysis results

Reply by: kbruff

Date:2/19/2004 4:51:37 PM

Rednroll said:
"Yes, I do this all the time. You'll need a program like Spectra or Smaartlive. The Sound Forge SA is lacking in that department, which means it's virtually useless."

Hello Rednroll,
Could you elaborate on how you use and why you use Smaartlive. As a novice to advance audio applications I did some research on the items that you spoke about. However I was interested in your application of spectral analysis, and average frequency response.


Also you mention TC Electronics Finalizer --how does this machine enhance your work flow and your projects?

Thanks in advanced,
Kevin
~~~

Subject:RE: Meaningful SpecAnalysis results

Reply by: farss

Date:2/20/2004 12:40:21 AM

I'm a bit of a newbie to all this but I am under the impression that Sonic is correct. If I highlight a section of audio on the T/L and with RT Update disabled in SA then hit refresh it shows me the average value of each band over the time period I have selected on the T/L. From memory it will also show me the peak value of each band over the sampled region as well.

But I do have a related question, can anything in SF do a decay analysis? By that i mean a 3d plot of x=frequency, y=amplitude and z=time.

Subject:RE: Meaningful SpecAnalysis results

Reply by: Sonic

Date:2/20/2004 8:37:40 AM

OK, I think we are just confusing what "average" means. You are talking about the equal power per octave "average"-ness of pink noise and it's usefulness in testing speaker response or what-have-you.

I am not talking about the excitation signal at all. I'm just saying that the overlapped analysis of a selection many times the length of the FFT is the "average" frequency response for that entire selection.

So I expect your drastically different snapshots involve no selection and are just the N-point FFT from wherever the cursor is positioned. Even with pink noise, this type of analysis will yield marginally similar spectra, but never identical, and sometimes very different depending on the noise content and the length of the FFT.

There are two approaches that will help. First, increase the length of the FFT to the maximum you can tolerate. Second, compare selections of longer duration (say 1 or 2 seconds) rather than individual FFTs at random spots.

Simple experiment:
a) generate 5 seconds of mono pink noise
b) set FFT length, 2048 is adequate for this
c) view the analysis just by clicking around with no selection, the frequency characteristic isn't obvious and changes a lot.
d) now start making random 1, 2, and 5 second selections and note that the analysis now approaches the expected smooth 1/f characteristic

Basically, the longer you make the analysis or selection, the more accurate the "pinkness" of the noise component gets. You can run the same experiment with white or brown noise and you'll see that it takes a good chunk of samples to accurately show the expected noise spectrum.

Am I getting warmer?

J.

Subject:RE: Meaningful SpecAnalysis results

Reply by: Sonic

Date:2/20/2004 8:41:41 AM

Sonogram mode is x=time, y = frequency, and color depth = freq. amplitude.

In normal mode, you can increase the number of "slices" in settings to get a waterfall style plot of equidistant single FFTs through the selection.

J.

Subject:RE: Meaningful SpecAnalysis results

Reply by: Rednroll

Date:2/21/2004 7:51:20 PM

"Could you elaborate on how you use and why you use Smaartlive. As a novice to advance audio applications I did some research on the items that you spoke about. However I was interested in your application of spectral analysis, and average frequency response."

I have a few different applications that use this type of functionality. The first is for mastering audio. I will use a reference CD track of a similar type of music that I am mastering that I know sounds good. I will therefore play the reference track with a realtime input SA that has an averaging function. I will take a snapshot of the spectral curve. Then I will start mastering the track by using the TC Finalizer, where I will use mainly the multiband compressor, EQ, and limiter. I can adjust the multiband compressor and EQ and have a visual representation of how the song I'm mastering compares to that of the reference audio spectral curve. When you start to add multiband compression for loudness it will alter the frequency curve, where I will then adjust the EQ to compensate for this. It's very easy to pinpoint and narrow in on the frequencies that need to be adjusted. I also use this similar process when mixing within Vegas. So for both of these applications, it is not practical to be able to select a portion of a wave file and get the spectral average curve, it must be a realtime averaging function. Otherwise it would be very time consuming recording and rendering part of the audio to be able to see the average curve within Sound Forge and you can't do realtime adjustments and immediately see the results.

The next use is for seeing the the spectral curve response using pink noise of an acoustical environment. So you set up microphones and monitor the spectral curve in realtime to see where the environment is causing major peaks and notches in the spectral curve of the pink noise. I am an engineer for Harman International that supplies premium audio systems in vehicles. Some of the brand names you might be more familiar with are JBL, Infinity, Harman/Kardon, and Mark Levison. A vehicle causes many obstacles for an audio system. Like you can't move the speakers, there's many reflective surfaces like glass, there's many trim parts that may resonate at certain Low frequencies, you can't add absorbant material where you want, road noise is particularly problematic for masking the low end response and there are also seats, seatbelts and map pockets sometimes located infront of the speaker. All of these things cause havok on the spectral content of the audio. So the first thing to do is play pink noise through the audio system and see where these problem frequencies peaks and notches are. For the system we have developed 6, 8, 10 and 12 channel digital amplifiers. Each channel has it's own EQ and delay adjustments. Sometimes just boasting EQ alone will not fix a notch, because the notch is happening because of phase cancelization of that particular frequency, so you can actually add a gain in the EQ and have it phase cancel more. So we will add delay to certain channels to elliminate that notch, which may cause another notch in another band. So it's a constant balancing act and playing around with different settings to try and ellimate all the problem frequency areas. So again, you need a realtime SA with an averaging function.

A similar application is doing the same thing when tuning the acoustical environment for my studio. You play the pink noise and do the same thing to see where the problem frequencies are. I prefer not to use EQ on my monitors to correct these problems. So what you do is while you have the realtime SA monitoring function going, you start to hang absorbant material on the walls, and you can see the SA in realtime, to locate the exact placement to hang the absorbant material to correct the problems. So again, for this it would not be practical to keep recording the pink noise signal into sound forge and getting an average curve because it would be very difficult.

I don't use Smaartlive, since I'm just more familiar with Spectra at this time and it does everything I need it too. Some of the other engineers that I work with prefer Smaartlive over Spectra, I'm really not sure why. I think one reason is because it has a VU meter along with the SA to see the loudness also.


"Also you mention TC Electronics Finalizer --how does this machine enhance your work flow and your projects?"

I think I answered that above. I find the sonic sound quality of the EQ's and multiband compressor far superior than any plugin I've used and again it's another familiarity thing and doing what I need it too.

red

Subject:RE: Meaningful SpecAnalysis results

Reply by: kbruff

Date:2/21/2004 10:28:09 PM

Thanks -- the information provided is very interesting.
I am an Electromechanical Engineer -- I have some awareness for Audio but there many ideas mentioned which create deep curiousity.

Sincerely,
Kevin
***