I just honestly don't think its as good of a benchmark as many people are giving it credit for. For example, the remastered Rush - Moving Pictures album gets a -5.13 dB. This is an album from 1981 simply remastered. I think it sounds very good and the special effects on songs such as Tom Sawyer and Limelight come out very well. But yet, it gets a -5 dB reduction, thus many would warrent it "too loud." My main example to support my view is Tool - Lateralus. Arguably one of the best mastered CDs in the last decade. Besides being an HDCD, and, well, a Tool album, it sounds stunning. The treble is clear and concise, the midrange is solid, yet not elevated, and the bass is strong and pronounced, but not overwhelming. What a CD should be. Yet, the album, as a whole, gets -7.17 dB reduction. Nine Inch Nails - The Downward Spiral is another solidly produced album with a gritty sound one could come to expect from Trent Reznor, yet the album gets a -7.77 dB reduction. Maybe we should start taking other factors into account besides how loud an album is? Perhaps genre, producer, etc.

Maybe just because an album is "loud" doesn't make it poorly produced.

replaygain doesn't report to you the quality of the work. it simply tells how loud the recording is. you make your own decision on whether you want to run it through replaygain or not.

I'm simply commenting on the usage of replay-gain by forum members as a tool to evaluate the quality of production of an album.

Second that.
Look at Norah Jones "Come Away With Me" This recording is increadibly loud for the content it has - but it sounds wonderfull!

There are many other loud recordings just sounding extremely good.

Wombat

There are many more variables to consider when determining whether a recording was mastered well, than just an RG value. A value of -5db would tell you, yes its a very loud and compressed/limited recording, but it doesn't tell you how it was compressed/limited, nor does it comment on the quality of the harmonic balance, etc.

Thats the point!

Many people are just using RG info as a benchmark, and nothing more. Hence, a loud album is a bad album in their eyes.

I don't think many people consider -5db to loud. It's when recording's start to distort that we all get annoyed.

A ReplayGain value of -5dB is yet nothing to worry about. I've got many discs with that value of RG and most of them sounds great. It's RG of -8dB or -10dB when things start to sound quite bad. Though it all depends on how well/bad was the mastering done and how "loud" the music itself is.

RG values are a factor to consider, and i find them a useful benchmark for "judging" an album's quality. but not as a first thing.

when i get a new CD, i'll listen to it before i do a replaygain scan (which involves ripping it of course), so the first (and arguably the last) test of quality comes from my ears.

if said CD sounds bad, mediocre or terrible, THEN i run it through replaygain and see if it tells me something i didn't already know (having listened to the CD, i've obviously had to adjust my volume control to the album's loudness, so i'll have a fair idea of what the RG value will be. i'm only surprised when i get a positive gain).

RG was of course never intended as a quality test after all - it was intended to place all songs (albums?) at the same objective level, based on loudness contours and statistics etc.

it would be fascinating to have a tool that judged the subjective quality of an album. i wonder if psychoacoustics have advanced to the point where an algorithm can determine whether an album sounds good or not. (i don't just mean counting how many fullscale samples there are, or dropouts in extreme cases).

hmm. i would call a RG value of -5 quite tame. Electric Six's "Fire" comes in at around -10, has evil clipping all over it, but somehow manages to sound pretty good to me - i guess it could be said that smashing fullscale was an artistic decision, but in reality it was probably a fluke.

at the moment i'm working for a small radio station, trying to make recordings of varied quality (from studio recordings to a minidisc placed on a table in a tiled kitchen) sound both intelligible and as loud as possible. there's no broadcast processor box to make everything loud, and i now have a new found respect for any mastering engineer who can punish a recording so severely and still get a good sound out of it

yeah. The only thing RG will tell you is if the dynamics are probably squashed or not.
Often out of interest I'll toss the song into a wave editor and see what's happening. But sometimes people can manage to make squashing sound good, even.

What immediately came to mind is how it would rate the "balance" of the song (bass vs midrange vs treble). Lots of songs work because they are _too_ bright or _too_ warm or _too_ bassy, and I wonder if an algorithm can tell the difference between a good unbalance and bad unbalance. Compression can also be used to make a song more exciting (via "pumping" artifacts & distortion)... how can it tell the difference between good dynamics and bad dynamics? etc.etc.

If you've ever done music mixing in general you'll know there's a really thin line between good (or exciting) and irritating. And even that will vary from person to person

An example :
Indochine - Unita - L'Aventurier

RG track gain : -11.55 db
Sounds good and quite dynamic ! An inspection in a wav editor shows that the peaks are clipped, but it is not audible. The high pitched guitar and percussion sounds must boost replaygain correction, because they fall into the sensitive part of human audition.

I think what is being done is comparing apples to oranges.

1) Replaygain tries to estimate the _perceived loudness_ of tracks in order to equalize perceived loudness of tracks to be somewhat similar. This is aching to trying to simulate one part of hearing (loudness perception) through an algorith by doing a very simplified calculation (e.g. it lacks temporal integration analysis, imho).

2) Mastering dynamic headroom or amount of dynamic compression is a different calculation. This can be done with various ways: comparing the original to the mastered version (algorithmically) or just comparing the mastered version to the limits of the carrying medium (such as cd dynamic range).

These two comparisons do not always correlate and as such it can be misleading to talk about them interchangeably.

Also, as for "quality of the mastering", can also be divided into two components:

A) Signal theory quality: are the dynamics compressed? Are the peaks limited? That is, is it in general distorted? If it is, it gets a worse grade. This measurement can be done with a meter (algorithm) if the source and the final engineered mix can be compared.

B) How does it sound to a person (with all artistic intentions included in evaluation)? Does it sound _too_ compressed for it's own good? Are the peaks clipped in a way that makes it just sound bad in _wrong way_? This evaluation can only be done with ears, there is no objective measure (replaygain or anything else) to assess this.

Again, B) is the only way to evaluate "the quality" of the engineering on a CD in a sense that the original poster wants. One can further divide B to a fan's POV and the artist's POV. These don't always match.

Replaygain values don't necessarily have anything to do with this.

regards,
halcyon

PS It has already been discussed elsewhere that Replaygain is not a reliable indicator of mastering stage dynamic compression (because it wasn't built to be such): http://www.hydrogenaudio.org/forums/index....topic=22187&hl=

There's also the fact that ReplayGain is based on an equal loudness contour, which means that, as mentioned before, it's designed to measure how loud the album SOUNDS, rather than how loud it's actually tracked at. Two songs with very differently EQ'd mixes with the same RMS and same song arrangement cna have widly different RG values. A song that's very dominant in upper-mid frequencies can have a value of -8 and still sound dynamic, wheras a bass-heavy song with that same value will more than likely sound rather squashed.

Case in point: Live's 1994 CD Throwing Copper has ReplayGain values of up to -9db, however the actual song RMS values tend to hover around 12db and the CD still manages to sound good because the mix is very dominant in midrange frequencies which tends to give it higher RG scores. Wheras Rush's Vapor Trails, an album who's mix is more low end-oriented, barely goes over -9db yet the sound is undeniably smashed to hell and back, with song RMS values around -9db.

There should be an option to turn off the loudness contour in Replay Gain so someone can get a more accurate analysis on the actual tracking level of the sound rather than just how the ear will percieve it.


It also has quite a bit of digital clipping on it, though.

-5dB is hardly "squashed." Since Replaygain is correcting to an average RMS of -20 dB
(in my experience), a replaygain of -5dB means an average RMS of -15dB. Bob Katz, a mastering engineer who is very outspoken against the loudness race, recommends a mastering level of -14dB average for this sort of music.

With a good limiter, recordings can go up to an average level of -12dB (Katz recommends this level for low-dynamic-range mediums such as broadcasting) without much noticable squashing, approximately equivalent to a replaygain of -8. Beyond that, recordings start to have a quite noticeable lack in punch and transient clarity (i.e. drums start to sound like boxes). At around -10-8 dB RMS (approx -10-12dB replaygain), you start hearing audible artifacts such as distortion, clipping, and pumping. Any louder than that is simply unacceptable to even the most casual listener.

This thread raises a question:

As it has been proven that RG is not a good indicator of mastering level hotness/compression/dynamic headroom, then what tool should people use?

Before you answer, there are several issues to be tackled (non-exhaustive list):

- headroom (average power to peak difference): this usually signifies how much modern pop music is compressed/peak limited (although it can be misleading)
- footroom: (noisefloor vs lowest musical power)
- dynamic range: the usable dynamic range for the average power of the music to change within (headroom is reserved for transients, footroom for lowest parts)
- number of clippping samples (as per CD's > 0 dB FS)
- how does it sound? Does it sound overly hot? Distorted? Congested? (Even though the instruments themselves by default are not like that when used normally?) This last one probably can't be measured at all (i.e. one needs to listen).

I don't know what the right tool is. Of course the normal average RMS power calculations are what most of studios seem to use.

Are there better tools? Free tools? Accurate tools?

Their ears?


Anyway, didn't we have this thread already?

Cheers,
David.

You are of course right, we did

And ears are best for perceptible quality, of course.

However, maybe I got the discssusion of to a wrong start.

I meant a technical (algorithmic) measure that calculates dynamic headroom, CD full scale clipping and many other issues into a single signal quality measure.

Something that does not try to assess the perceptual quality of the compression, but to assess it's effects on the signal itself. To me AVG RMS power is not all encompassing.

Perhaps such a measure is not needed, but I'm curious nevertheless

I'm not so well versed in various studio mastering / engineering metering techniques, so maybe a combo like this already exists.

If anybody knows of single or compound measures, please chime in.

There's the basic problem. As long as people continue to refer to the RG settings as "scores," unknowledgeable people that are just listening in will assume a value judgement.

RG is no different than BPM -- it's just a number.

-- Rick