[TenTec] RF Speech Processor

[Pete Ferrand]

Maybe this will help to understand a little bit about processing:

First of all, the normalize function brings the highest peak of the audio up to the maximum level specified by the user. So if you normalize to 99% then the highest peak will be at 99% and everything else will be at the same relative level as it was before the normalizing. Running the normalize function again at 99% will make no difference. Normalization has nothing to do with the average value, although of course the average value does increase, if the normalization process increases the peak value to the 99% in my example. If the peak value is 100% originally and you set the value at 99%, then everything will be reduced so the peak is at 99%. The normalization process does not address the "average" at all, the average just changes as a result of where it puts the peak.

If prior to the normalization you have one peak at 95% and the rest of the audio at about 20%, then normalizing to 99% will increase the peak to 99% and the rest of the audio to 24%.

A compressor does not fold energy from outside the frequency spectrum. It has nothing to do with frequency whatsoever. There are some compressors that are "split band" that compress at different rates for different portions of the frequency bands but nothing ever gets folded from one part of the frequency spectrum to another. The compressor works only on amplitude.

Now, in my 40+ years of working with audio processors in broadcasting, I can say that every modern radio station uses multiple audio processors on the audio. All of these compress to some extent, although they have different names such as "volume leveler", "limiter", etc. Now that we have DSP, these different audio processors are incorporated into one physical box.

You certainly can, and should, compress the signal multiple times for maximum punch and maximum sound quality. Each compressor is designed to distort the signal in different ways so it will sound better and overcome various limitations of band and equipment. A recording studio may have dozens of compressors, some hardware, some software, and frequently used in series, depending on what they're trying to do.

If you look at the controls on a recording or broadcast compressor they all have settings for attack time, release time, and input gain. Some have other controls such as where the "knee" is, makeup gain, and so on.  The knee is, to oversimplify, the volume point at which the compressor starts compressing. This may be a hard knee where it starts compressing heavily at a certain point, or a soft knee where the amount of gain reduction increases gradually with increasing volume.

In the old days of the 40's and 30's the compressor consisted of a bored engineer with his hand on a volume control. However this function became enhanced with processing designed to bring up low levels by reducing the gain of high level signals. These functions call for a slow attack and release time. A sharp peak on a voice or snare drum would still be missed by the device and overdrive the transmitter, so limiters were introduced which cut off the maximum peak at the 100% point or thereabouts. However you don't want to do this a lot because it will sound terrible so in between the two you have a compressor with a relatively fast attack and some middle release time so the limiter is rarely invoked. 

In some cases we add an expand function to undo some of the compression.  For instance, you can take the compressed audio and expand ("uncompress") from, say -15 to -25 dB if you want. Then we can set up another compressor that avoids compressing the lowest level sounds so room noises and fans aren't brought up to voice level. We can also have the compressor only look at part of the audio frequency spectrum, so bass notes don't cause compression to increase over the entire frequency spectrum. 
My point is that all compressors have a lot of parameters and after one compressor has its way with your voice, another may do additional processing to accomplish something else. No one set of parameters will do an optimum job of keeping the average level up while sounding natural. You can run audio through four compressors and still have a natural sounding voice, but it will cut through better. Setting up compressors takes both knowledge and trial and error and is very controversial as this thread indicates. 

An analogy might be in Photoshop type processing. You have brightness and contrast, as you only have so much dynamic range in a photo that the display or printer can handle. But you also have saturation, gamma, thresholds, and many other things, all of which interact. As with sound, there is no one setting that will bring out what the myriad of settings will.

Is any of this necessary? No, but the technology is there if you want it. 

Hope this helps.

-Pete
WB2QLL
Somers WI








>	Yes... but the energy is different. Once you
>	smooth it out, I am not sure you can smooth
>	it again.
>
>		Here is my theory:
>
>	A compressor sorta folds energy from outside
>	the operating frequency spectrum into the
>	useable spectrum.  (An admittedly inaccurate
>	description, but I am not working on my thesis...)   ;-)
>
>	The net result is that the average amplitude of
>	the signal and hence average transmitted power
>	ends up stronger than it would be had compression
>	not been used.
>
>	My guess is that you cannot "average" it twice.
>	Once it is averaged, I figure you cannot do it again.
>
>	It is analogous to running the "normalize" feature
>	in an computer audio processor application, which
>	is the application of a constant amount of gain to
>	an audio recording to bring the average or peak
>	amplitude to a target level (the norm).  Because the
>	same amount of gain is applied across the given range,
>	the signal-to-noise ratio and relative dynamics
>	are generally unchanged. Normalization differs
>	from dynamic range compression, which applies
>	varying levels of gain over a recording to fit
>	the level within a minimum and maximum range.
>
>	Now I figure once you have normalized it, you
>	cannot really normalize it again.   Similarly,
>	once you compress it, you cannot really compress
>	it again -  like, ...er... um... ah...  if you
>	get a hair cut, then you cannot get another cut
>	that same day without ending up with pretty short
>	hair...  once you cut it to a new length, you cannot
>	cut it again to that same length, and similarly you
>	cannot compress the signal twice - once you cut
>	the rough edges, it is smoothed out (averaged)
>	already.   (Sheesh... I wish I could think of a
>	better analogy... but I hope you get my drift...)
>
>	Parenthetically, those Omni VII settings sound
>	just about right.
>
>
>===================  K8JHR ====================
>
>
>
>
>'.
>
>	
>
>
>_______________________________________________
>TenTec mailing list
>TenTec at contesting.com
>http://lists.contesting.com/mailman/listinfo/tentec