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: Announcing my website: LoudnessRace.net (Read 9173 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Announcing my website: LoudnessRace.net

Reply #25 –
Quote
A small guide how to determine the main "loudness" problems of cd's may be helpful.
Are there any characteristic values? E.g. peak amplitude, percentage of clipped samples, rms-power, something like that?

A high RMS power would suggest that limiting and/or dynamic compression has been applied resulting in equally (too) loud sound all over the track.

Clipped samples can't be found, only samples with max/min sample vallue assuming that these sample values suffered clipping. But what if amplification by e.g. -0.2dB was applied after clipping had been caused? Then we never see max./min. sample values (32767 / -32768) but e.g. lots of values arround +/- 32021 ('arround' if dither had been applied as a last step). Or what if hard limiting has been applied without causing clipping? Then we see a big percentage of sample values e.g. between +/- 32000 and +/- 32767.

To get somewhat comparable numbers about peak values, we can amplify in a 1st step (without dither!), e.g. by 0.5dB resulting in all sample values between +/-30935 and +/-32767 becoming +/- 32767 (=clipping). The 2nd step would be counting +/- 32767 sample values (can also be done using a hex editor) and calculating the percentage of their occurance.
Let's suppose that rain washes out a picnic. Who is feeling negative? The rain? Or YOU? What's causing the negative feeling? The rain or your reaction? - Anthony De Mello

Announcing my website: LoudnessRace.net

Reply #26 –
Quote
I have a related question:

AFAIK for getting louder sound without too bad distortion (clipping etc.) limiters are used that work e.g. like this if I understand correctly:

IF -16000<= Samplevalue <= 16000 Samplevalue_New:=Samplevalue * 2
ELSE IF Samplevalue > 32000 Samplevalue_New:= 31360 + Samplevalue * 0.04
ELSE Samplevalue_New:= -31360 + Samplevalue * 0.04

No, it's a dynamic, varying thing. Imagine a volume control which you manipulate to increase the loudness during quiet moments, and reduce it during loud moments. Now automate the process, so you can respond instantly to loud peaks. Now split the signal into several frequency bands, so each frequency range can be made as loud as possible. That's basically the multi-band compression process, used on radio and many CDs. If there are any peaks left, you can compress these using a similar process, or you can just slice them off. Bingo - one loud CD ready for release in 2003.

(You can do what you said, but it adds distortion. Not that the above process doesn't, but used carefully, and without the final slicing, it wouldn't)


Quote
If 'real life' music is transformed to frequency domain, it's obvious that peaks (in time domain) consist of an addition of several sine waves. A different approach to lower the peaks would be to phase shift some of these sine waves and/or to add short, very high or very low frequency (mostly inaudible/masked by original signal) fade-in/fade-out tones. Both would need to be calculated in a way that peaks are lowered, therefore amplification without limiting or clipping is possible. (I hope it's understandable what I mean.)

Is this possible - or does it even already exist?


Yes, it's used in optimod radio processors - especially the AM ones.

http://www.orban.com/

Cheers,
David.

Announcing my website: LoudnessRace.net

Reply #27 –
Quote
Clipped samples can't be found, only samples with max/min sample vallue assuming that these sample values suffered clipping. But what if amplification by e.g. -0.2dB was applied after clipping had been caused? Then we never see max./min. sample values (32767 / -32768) but e.g. lots of values arround +/- 32021 ('arround' if dither had been applied as a last step).

well, strictly speaking, yes. But as CEP does, you can use a relative comparison to the peak volume in the track itself to find 'clipping' even though there strictly aren't any samples pegged to full scale.  So a last step of normalizing to -0.2dB isn't going to trip this method up, and you will still be able to find samples that are pegged to the relative 'full scale' value, and that's exactly what CEP does.  Jane's Addiction's Strays album has some particularly bad examples of clipping in it. Some of the quieter portions of the songs have obviously been limited past full scale and contain clipping, and were later lowered in volume to put them at the desired relative volume in the song. If you zoom in on one of these quiet regions you can see tons of flatspotted peaks even though the loudest samples in said region are nowhere near full scale.  Consequently, after running 'clip restoration' on the whole song those quiet sections were still full of clipping because of their relative position to full scale. I still successfully used 'clip restoration' on these quiet regions simply by selecting the region in question and applying the filter again to that section only. It used the peak value from the selection only, and 'rebuilt' the peaks just fine.

Announcing my website: LoudnessRace.net

Reply #29 –
Quote
Good website. The best part about it is that it makes people aware of the issue. Many people seem to take it for granted, and that's exactly the attitude that allows this race to keep on keeping on.

Also, i hate to say it, but your "2005" graph has already become reality. Recently i wanted to copy the album "St. Anger" by Metallica (private copy, for all you industry people) and wondered if my drive was broken when i looked at the mess in Cool Edit. But take a look for yourself, i.e. song #2. The whole thing is hard-limited shortly below 100%, everything you see is non-fullscale clipping.

I've heard St. Anger.  Bad as it is, it's nowhere near the 2005 graph.  The clips I took were from a pop song (Tears for Fears - "Everybody Wants to Rule the World") and had RMS values as follows:

1985: -18db
1990: -15db
1995: -12db
2000: -9db
2005: -6db

My analysis of "St. Anger" gave an RMS avg. of -9, which was the equivalent of the the 2000 example.  The loudest commercially-promoted CD I've heard would be either Andrew W.K. - Girl's own Juice EP or Queens of the Stone Age - Songs for the Deaf, both of which barely break -8.

As bad as commerical CDs are, they still have a ways to go before hitting some of the levels shown by independant noise releases, not to mention the 1997 remaster of Raw Power, which goes as high as -3!

Quote
Overcompression is going to be, at times, a subjective thing. What conditions must be met before the CD is considered to be overcompressed? I ask since it could be the way it was intented.

I consider -12db a good baseline for the average pop track, however this is only taking into consideration the majority of commercial rock music.  Obviosuly, something like Dave Matthews Band would give lower results and, say, Fear Factory higher results for the same percieved loudess, not to mention the fact that long periods of extreme dynamics can also throw the averages off.

Once you go past -11 the quality will start to audibly suffer, although it's not really until you hit -9db that things really start to become unlistenable.  I don't consider myself an audiophile, however, and sometims I find their more extreme views very offputting (I once read a letter written by a guy who considers anything released after 1993 "ruined".  Oooooookay...).

Announcing my website: LoudnessRace.net

Reply #30 –
Quote
Queens of the Stone Age - Songs for the Deaf

That's so ironic.


"You have the right to remain silent. Anything you say will be misquoted, then used against you."