Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: "Loudness War" article could use some informed comments (Read 10025 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

"Loudness War" article could use some informed comments

Reply #25
I don't mind the use of mp3s to measure RMS - the problem is that the masters iTunes have probably won't be the originals.


Interesting comments - DRC to make things listenable in the car needs to reduce the "macro" dynamics greatly - but most pop music is naturally like this anyway. It would be mostly listenable in the car without any DRC! For classical, reducing the dynamics to work in a car makes the recording sound a bit naff at home IMO. OK for background music, but a bit bland for really listening to.

Loudness war-style DRC reduces the "micro" dynamics too - in a way which doesn't help at all in the car or at home IMO. Those small and/or quick dynamic changes can stay intact even in the car - and the music really benefits! Shame it doesn't happen much.

Cheers,
David.

"Loudness War" article could use some informed comments

Reply #26
Compression is required as line level would wreak havoc on just about any system, mobile or not.  In a church service with an i-made-it-up 105db dynamic range, compression is required.  However, the engineer doesn't have to compress it to 3db of dynamic range.  There's 80db available on CD.  Squish it to 40 and it'll still sound good.

"Loudness War" article could use some informed comments

Reply #27
FLAC is open source, but the implementation is also monetarily free, so doesn't that count as "free?" Seems like the "open source is not free" topic is another discussion.
I would just call it FOSS, then http://en.wikipedia.org/wiki/FOSS


"Loudness War" article could use some informed comments

Reply #29
Do we really need to continue the off-topic nitpicking?



Good question.

As to the article, well, I read it. The author mucked up the description of perceptual coding to the point that I can't recognize it, which I think says something. I've actually pinged some of the NPR engineers, who very much know better, and we'll see what, if anything, comes of it.  I suspect journalism is involved, Bob Ludwig is certainly not so silly...
-----
J. D. (jj) Johnston

"Loudness War" article could use some informed comments

Reply #30
Lossy compression works by reinterpreting music as a collection of tones. The tones that are naturally masked and thus not recognized by the ear/brain are removed. When we decode a lossy-compressed file we don't notice the tones that were removed because we could never hear them in the first place.  Removing the portions of the music we never hear anyway vastly reduces the amount of data it takes to represent the music in a way that is pleasing and recognizable to our ears.

Your description is definitely easier to parse. Yours is a description I've heard before and it led me astray. It is not generally true to say that masked tones are removed. Masked frequency regions are coded at lower resolution (i.e. higher noise). Only if a masked tone is of sufficiently low level, and coded at sufficiently low resolution, will it be coded as a 0 and thus removed. Coders remove detail from masked areas (replacing it with a low-bandwidth approximation of  the original signal). In most cases, we're not wholesale nulling portions of the spectrum.

"Loudness War" article could use some informed comments

Reply #31
As a relative noob, I'd like to ask how the RMS explanation is off?  In particular, I thought the peaks and valleys picture was very informative.

RMS is the roughly the average level of the signal. The RMS level would approximately correspond to what is marked as "valley" in the diagram. But it is not reliable to visually measure a signal like this; you need to do the math.

Peak level is correctly marked. The distance between peak and valley is roughly the dynamic range or more accurately the crest factor. It is not RMS.

"Loudness War" article could use some informed comments

Reply #32
RMS is the roughly the average level of the signal.
The RMS of 1,-1,1,-1,1,-1,1,-1....... is 1. The average is zero. They are not even roughly the same.
lossyWAV -q X -a 4 -s h -A --feedback 2 --limit 15848| FLAC -5 -e -p -b 512 -P=4096 -S-

"Loudness War" article could use some informed comments

Reply #33
Perceptually lossly compression (MP3, AAC, DTS, etc) works by analyzing the audio signal and discarding the inaudible and least audible parts.

Any more "interpretation" is simply wrong. It's not broadband quantization, it's not capturing a tone here and there, etc, it's simply analyzing a signal for the inaudible and then (when you run out of bits) least audible parts, and discarding them.
-----
J. D. (jj) Johnston

"Loudness War" article could use some informed comments

Reply #34
RMS is the roughly the average level of the signal.
The RMS of 1,-1,1,-1,1,-1,1,-1....... is 1. The average is zero. They are not even roughly the same.

Audio signals do not have a DC component so all audio signals have an average value of 0. Sorry if the meaning of average level was unclear. By this I meant the average absolute value which is roughly the RMS level.

"Loudness War" article could use some informed comments

Reply #35
Audio signals do not have a DC component so all audio signals have an average value of 0. Sorry if the meaning of average level was unclear. By this I meant the average absolute value which is roughly the RMS level.

There is absolutely nothing that says that the average value in PCM must be 0. In fact, 8-bit PCM is unsigned, so the typical average value is 128, and an average value of 0 would only occur for dead silence.

What you probably meant to say was the average absolute value relative to the DC level of the signal.

I believe that the only signal whose average absolute value is equal to its RMS is a square wave.

"Loudness War" article could use some informed comments

Reply #36
Audio signals do not have a DC component so all audio signals have an average value of 0. Sorry if the meaning of average level was unclear. By this I meant the average absolute value which is roughly the RMS level.


How can that be since the average is computed as the sum of the squares.  For 1's that works, but that's about it.

The waveform is also a factor.  For a square wave, the RMS value is the absolute value of the peak volatge.  For a sine wave, it is the absolute value of the peak voltage / the square root of 2.  The details are found here in the "RMS of common waveforms" section.

"Loudness War" article could use some informed comments

Reply #37
Audio signals do not have a DC component so all audio signals have an average value of 0. Sorry if the meaning of average level was unclear. By this I meant the average absolute value which is roughly the RMS level.


How can that be since the average is computed as the sum of the squares.  For 1's that works, but that's about it.

The waveform is also a factor.  For a square wave, the RMS value is the absolute value of the peak volatge.  For a sine wave, it is the absolute value of the peak voltage / the square root of 2.  The details are found here in the "RMS of common waveforms" section.



I think there's quite a bit of technical confusion going on here.

First, an L1 norm is not an L2 norm, as the immediately prior poster points out.

Second, an L2 norm produces the POWER of the signal, and only the power. It does not address what we really care about in this thread, which is LOUDNESS.

Loudness is not power.
Power is not loudness.

Loudness can vary by a factor of very nearly 10 for a constant power, given comparing a single tone to a very unpleasant multitone.  A change of 10 in loudness is a very, very large, one might even say enormous, change.  That value is, of course, extreme, and an edge case.

Loudness for something that has the same spectrum, but only power changing, varies as the 1/3.5 POWER of the power.

There is a loudness tutorial deck at www.aes.org/sections/pnw/ppt.htm that has a lot more to read.
-----
J. D. (jj) Johnston

 
SimplePortal 1.0.0 RC1 © 2008-2021