IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
Multiband ReplayGain? Does it make any sense?
darkbyte
post Sep 12 2013, 11:31
Post #1





Group: Members
Posts: 145
Joined: 14-June 11
Member No.: 91517



Current ReplayGain implementation calculates a global loudness value. I'm wondering if it makes sense to create a new variant which calculates ReplayGain values for subbands? Say i like to make the equalization constant, a player which understands these new gain values could automatically adjust the EQ values to match my needs. A track which already contains loud trebles will not get any treble boost and vica versa. I know multiband compressors are doing the same in real time but this wouldn't require anything special during playback since it's only a matter of adding an EQ based on the precalculated values.

This post has been edited by darkbyte: Sep 12 2013, 11:35


--------------------
Wavpack -b450x1c
Go to the top of the page
+Quote Post
skamp
post Sep 12 2013, 14:17
Post #2





Group: Developer
Posts: 1412
Joined: 4-May 04
From: France
Member No.: 13875



Earlier thread on that topic: Feeding an EQ to a Replaygain scanner


--------------------
See my profile for measurements, tools and recommendations.
Go to the top of the page
+Quote Post
darkbyte
post Sep 12 2013, 17:08
Post #3





Group: Members
Posts: 145
Joined: 14-June 11
Member No.: 91517



Ah, i've remembered we've had a talk about this earlier. smile.gif Although you're idea works differently because it still calculates one replaygain value, not sepearate values for each band of the original file.

My question from that topic is still valid though. I wonder what happens if i bandpass the spectrum into different bands and apply the replay gain RMS calculation on top of them. Do i get usable gain values for a further EQ correction of the bands?

This post has been edited by darkbyte: Sep 12 2013, 17:18


--------------------
Wavpack -b450x1c
Go to the top of the page
+Quote Post
pdq
post Sep 12 2013, 17:24
Post #4





Group: Members
Posts: 3377
Joined: 1-September 05
From: SE Pennsylvania
Member No.: 24233



Somehow, making all of the bands equally loud doesn't sound like a good idea. Or am I misinterpreting what you wanted?
Go to the top of the page
+Quote Post
darkbyte
post Sep 12 2013, 17:39
Post #5





Group: Members
Posts: 145
Joined: 14-June 11
Member No.: 91517



Say i like the sound of an EQ preset where the bass and highs are boosted. Some songs have already very loud bass and treble, so it's not a good idea to boost them further because it will introduce distortion and/or heavy clipping. In this case the player uses an offset calculated from the band's replaygain to reduce the effect of the EQ at those bands. Some songs may have very quiet mids and highs and very loud bass (The Fugees: Ready or Not is a very good example for this). In this case the player knows that it has to raise the gain on the mids and highs to keep the overall EQ image constant.

This thing is very close to what radios are doing to have a constant sounding of the station. It's not something you have to use all the time. If you wish to return to the original mastering you just simply turn of the correction. It's only an additional metadata, nothing's altered. But if you want to listen to radio like sound or hate altering the EQ from song-to-song this might be a solution if you don't want to use heavyweight DSPs.

This post has been edited by darkbyte: Sep 12 2013, 17:40


--------------------
Wavpack -b450x1c
Go to the top of the page
+Quote Post
db1989
post Sep 12 2013, 17:40
Post #6





Group: Super Moderator
Posts: 5275
Joined: 23-June 06
Member No.: 32180



On the simplest level, affecting the sound in any way other than adjusting its overall volume goes far beyond the intent of RG, R128, and co. Sure, some of the algorithms assess perceptual loudness in ways that consider different frequencies/bands separately, but those considerations are collated to adjust the final loudness only.

Also, as pdq pointed out, there are likely to be logistical issues implementing what you want in any form. Tandem to this, it would help if you would make your hopes clearer. What do you actually want? Something that can take an EQ curve of your preference and adjust a given track on a multi-band level until its own overall spectrum approximates that curve? What sort(s) of windows should it use? There are likely to be a lot of big questions like this involved.

In any case, I doubt ReplayGain is the right context for talking about this idea.
Go to the top of the page
+Quote Post
darkbyte
post Sep 12 2013, 17:46
Post #7





Group: Members
Posts: 145
Joined: 14-June 11
Member No.: 91517



QUOTE (db1989 @ Sep 12 2013, 18:40) *
What do you actually want? Something that can take an EQ curve of your preference and adjust a given track on a multi-band level until its own overall spectrum approximates that curve?

Yes, exactly. The difference is that multiband compressors are doing this in real time and adjusting their parameters very fast based on the input signal. My idea is to calculate an average loudness (RMS?) of the subbands during the whole track and store these values in the metadata and while playing apply a calculated EQ (based on the metadata and the user's preference) to the whole track. This is why i thought about ReplayGain which does the same but for the whole spectrum. Moving on to the next track will yield in a different EQ correction, and so on.

I have very little knowledge in DSPs so yes it's possible that my idea is wrong in it's roots but i was curious if it's doable or not and if it will sound natural in the context of radio like processing. I know that DSPs made for radio stations are a lot more complex than that. I don't want such heavy audio alternation but something that's lightweight enough to be even used on a portable device but still gives an acceptable result.

This post has been edited by darkbyte: Sep 12 2013, 17:59


--------------------
Wavpack -b450x1c
Go to the top of the page
+Quote Post
2Bdecided
post Sep 18 2013, 12:11
Post #8


ReplayGain developer


Group: Developer
Posts: 5066
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



The problem you hit with a simplistic implementation is that some tracks should have more highs or lows than some other tracks, which leads to some weird corrections being applied if you blindly match them. Compare a solo female voice with a typical dance track: you can't match the amount of bass on them.

I bet this is a solved problem (content type identification, gating, limited changes etc plus loads of knowledge from compressor design), but I haven't tried to find that solution. I end up manually remastering (mostly just re-EQing) the worst recordings I have to fit in with what I consider to be "normal".

Dynamic range, stereo separation, the amount of reverb and the amount of background hiss are also things that vary dramatically between certain recordings and could be automatically or manually "fixed" - but you're really heading into "easier to do more harm that good" territory when you mess with those, even manually. That shouldn't stop you doing it for your own listening if you really want to though.

Cheers,
David.
Go to the top of the page
+Quote Post
2Bdecided
post Sep 22 2013, 11:10
Post #9


ReplayGain developer


Group: Developer
Posts: 5066
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



Thinking about this more, I think you could use replaygain to do the matching for similar types of music. I'm imagining it as a tool within an audio editor. There should be one track or preset which you want to match to, and another track which is the actual audio you want to change. Then you pick the number of analysis bands and how much correction you want to apply (100% or less) and listen to the result. If you don't like it, try a different number of bands or different target eq or a different amount of adjustment.

Gated rms measurement is probably a good start. Ebu r128 is fine, but with enough bands the ebu r128 filter becomes irrelevant.

Someone has probably already solved the problem even more elegantly, but the O.P.'s suggestion is a good idea if you match the levels of each band to something sensible and relevant.

Cheers,
David.

This post has been edited by 2Bdecided: Sep 22 2013, 11:12
Go to the top of the page
+Quote Post
darkbyte
post Sep 22 2013, 11:54
Post #10





Group: Members
Posts: 145
Joined: 14-June 11
Member No.: 91517



Thanks David for explaining.
Your example of comparing dance music to a solo female vocal track is makes sense. I thought about this aswell. My idea is very raw in it's current form so i never thought this would end up in any useable in it's current state. But i thought it would be a great conversation starter for thinking about something similar smile.gif


--------------------
Wavpack -b450x1c
Go to the top of the page
+Quote Post
2Bdecided
post Sep 23 2013, 13:44
Post #11


ReplayGain developer


Group: Developer
Posts: 5066
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



With a good ear and experience, it would be quicker just to use EQ to get what you want - but your idea offers the possibility of semi-automation - useful for batch processing and/or when you don't have a clue what to do.

I've tried it on a couple of songs and it worked well. I need to try it on some more demanding and obvious tracks.

Cheers,
David.
Go to the top of the page
+Quote Post
2Bdecided
post Nov 7 2013, 12:34
Post #12


ReplayGain developer


Group: Developer
Posts: 5066
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



Somewhat related...

http://www.zynaptiq.com/unfilter/
http://www.youtube.com/watch?v=7thIB9VX35k

You can "unfilter" sounds, and apply the EQ/spectral balance of one signal to another.

They say they're using blind deconvolution to figure out the unknown filter.

Cheers,
David.
Go to the top of the page
+Quote Post
Specy
post Nov 8 2013, 14:11
Post #13





Group: Members
Posts: 22
Joined: 27-November 08
Member No.: 63320



It can be done but not with something like ReplayGain - your output will be a completely new file (which I would advice not to convert to MP3 again).

The trick for multiband compressors, that you could also use for this, is:
1 - Avoid huge differences between bands.
2. - Avoid very steep slopes between bands.
3. - Have a maximum amount of correction. If one band is decreased by 20 dB and another is increased by 20 dB, you can be pretty sure that it will sound horrible. See also the previous 2 points smile.gif
Go to the top of the page
+Quote Post

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 1st August 2014 - 00:27