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: Replay gain is not perfect (Read 11441 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Replay gain is not perfect

Well, Replay Gain helps a lot as it can actually correct the volume level across the tracks.

But still, this algorithm is far from being perfect. I think someone should take an investigation and write a completely new algorithm that would count much more deeply with psychoacoustics.

What I mean, is the Replay Gain behaves too strictly or better to say it is too limited, it is not clever enough to analyze the real loudness in the tracks.

Because, as a result when I encode a Soundgarden track, and then a track where a single guitar player plays a standalone guitar with some high pass frequencies effects - the standalone guitar sounds too loud. And a human ear can clearly recognize that the standalone guitar is really louder then Soundgarden. As opposed, human ear would expect Soundgarden as a bunch of loudy guitars to sound louder...

If you analyze the tracks in the wave editor, you can clearly see that the spectrum level of a loud Soundgarden tracks is set too low compared to standalone guitar track... this should not happen, the algorithm should put the volume level so that the frequencies appear more loudly in Soundgarden track I am pretty sure - in other words: it should recognize the critical frequencies that make the real loudness and put these frequencies at the same volume level that would make the human ear think that the traks are pretty much the same volume level even when thea are completely different as for the source of sounds

I think the Replay Gain does not corretly analyze the loud frequencies that human ear recoginzes as real loudness - and that is why it gives you the results that sometimes especially standalone instruments or voices are too loud compared to mixed instrumental sound.

I could post a test OGG files so that you could easily hear that Replay Gain is far from perfect.


I still admire it since it helps a lot anyway.

Replay gain is not perfect

Reply #1
Quote
What I mean, is the Replay Gain behaves too strictly or better to say it is too limited, it is not clever enough to analyze the real loudness in the tracks.

Because, as a result when I encode a Soundgarden track, and then a track where a single guitar player plays a standalone guitar with some high pass frequencies effects - the standalone guitar sounds too loud. And a human ear can clearly recognize that the standalone guitar is really louder then Soundgarden. As opposed, human ear would expect Soundgarden as a bunch of loudy guitars to sound louder...


I find this highly impossible. ReplayGain takes into account equal-loudness contours. What you are complaining about is more of mastering problem.  That transcends beyond what ReplayGain is capable of. You can blame shitty mastering engineer's with their multiband compressors in short.  This isn't a remaster btw is it?

Quote
I could post a test OGG files so that you could easily hear that Replay Gain is far from perfect.


... and btw they are Vorbis files is that so hard for people to understand? Thank god wikipedia is attempting to get people out of that habbit.
budding I.T professional

Replay gain is not perfect

Reply #2
RG doesn't change the relative volume differences within a track, and nor should it.

Replay gain is not perfect

Reply #3
Quote
RG doesn't change the relative volume differences within a track, and nor should it.
[a href="index.php?act=findpost&pid=355271"][{POST_SNAPBACK}][/a]


Correct, but the topic starter made no such claim.

Replay gain is not perfect

Reply #4
If I understand the poster claims correctly, what he means is that replaigain is making the solo guitar sound as loud as the soundgarden passage.

We could discuss if this is the goal of Replaygain, or if it should differentiate, like maiki says.

Probably, a way to differenciate this, would be the dynamic range left.

Replay gain is not perfect

Reply #5
We are aware the Replaygain isn't as accurate as it could be, but someone capable of tuning to perfection is hard to get. Applying Zwicker model and revising passes etc takes time along with knowledge.

It is possible it will be perfected in time, but then spawn from foobar2000 'branch' which is the only regularly updated implementation.
Efficiency is what has been worked on the most and reports on 800x+ realtime calculations have been reported on new systems.
Other improvements are channel handlig and other older limitations.

The one indivudual with the initiative to perfect the actual analysis process (former developer of Musepack - Frank Klemm) has little - none to be precise - involvement in this development today so there's not exactly and  ETA on when this would be worked on.

Considering the small factor of actual "errors" Replaygain has today, it has  a low priority flag. For the most common applications there is little to be noticed.

Replay gain is not perfect

Reply #6
You can lower the effects of these "inaccuracies" by using albumgain.
I am arrogant and I can afford it because I deliver.


Replay gain is not perfect

Reply #8
Quote
Hmm. I can't understand the original poster very well, but I was just about to recommend track gain.
[a href="index.php?act=findpost&pid=355282"][{POST_SNAPBACK}][/a]

he is complaining about quiet tracks being amplified so much that they seem louder than the loud tracks. He probably IS using trackgain.
I am arrogant and I can afford it because I deliver.


Replay gain is not perfect

Reply #10
Replaygain simply doesn't account for the fact that we expect solo guitar to be lower in loudness than a metal band playing at full volume.

It treats all audio the same, because it just sees bits and bytes, it is unable to distinct between a solo guitar and a metal band.

Replay gain is not perfect

Reply #11
Quote
Because, as a result when I encode a Soundgarden track, and then a track where a single guitar player plays a standalone guitar with some high pass frequencies effects - the standalone guitar sounds too loud. And a human ear can clearly recognize that the standalone guitar is really louder then Soundgarden. As opposed, human ear would expect Soundgarden as a bunch of loudy guitars to sound louder...

Do you mean you expect that the "relative loudness" between tracks are kept intact? Isn't this what "album gain" is designed for?

Edit: the sentence showing he is talking about 2 tracks is in bold.

Replay gain is not perfect

Reply #12
Quote
.. and btw they are Vorbis files is that so hard for people to understand? Thank god wikipedia is attempting to get people out of that habbit.

*habit
OT but I think it's silly to get in an uproar about something this minor.  Fact of the matter is, most formats here are referred to based on their file extension, including AAC, MP3 and MPC.

Replay gain is not perfect

Reply #13
Quote
Quote
.. and btw they are Vorbis files is that so hard for people to understand? Thank god wikipedia is attempting to get people out of that habbit.

*habit
OT but I think it's silly to get in an uproar about something this minor.  Fact of the matter is, most formats here are referred to based on their file extension, including AAC, MP3 and MPC.
[a href="index.php?act=findpost&pid=355392"][{POST_SNAPBACK}][/a]


I've been thinking the same thing lately. I see a lot of correcting people on the fact that it should be Vorbis instead of Ogg. But I never see people upset about references to WAV files when it should be PCM according to that logic. Technically this is accurate but it just seems extremely nitpicky to me.

Replay gain is not perfect

Reply #14
Quote
Because, as a result when I encode a Soundgarden track, and then a track where a single guitar player plays a standalone guitar with some high pass frequencies effects - the standalone guitar sounds too loud. And a human ear can clearly recognize that the standalone guitar is really louder then Soundgarden. As opposed, human ear would expect Soundgarden as a bunch of loudy guitars to sound louder...


Stupid question, but wouldn't using album gain have completely eliminated this problem?  I mean if you want to preserve intertrack dynamics, don't use the mode that removes them

Replay gain is not perfect

Reply #15
Quote
Quote
Quote
.. and btw they are Vorbis files is that so hard for people to understand? Thank god wikipedia is attempting to get people out of that habbit.

*habit
OT but I think it's silly to get in an uproar about something this minor.  Fact of the matter is, most formats here are referred to based on their file extension, including AAC, MP3 and MPC.
[a href="index.php?act=findpost&pid=355392"][{POST_SNAPBACK}][/a]


I've been thinking the same thing lately. I see a lot of correcting people on the fact that it should be Vorbis instead of Ogg. But I never see people upset about references to WAV files when it should be PCM according to that logic. Technically this is accurate but it just seems extremely nitpicky to me.
[a href="index.php?act=findpost&pid=355396"][{POST_SNAPBACK}][/a]

Exactly, and MPC should be referred to as Musepack, similarly.

Replay gain is not perfect

Reply #16
Quote
Quote
Because, as a result when I encode a Soundgarden track, and then a track where a single guitar player plays a standalone guitar with some high pass frequencies effects - the standalone guitar sounds too loud. And a human ear can clearly recognize that the standalone guitar is really louder then Soundgarden. As opposed, human ear would expect Soundgarden as a bunch of loudy guitars to sound louder...


Stupid question, but wouldn't using album gain have completely eliminated this problem?  I mean if you want to preserve intertrack dynamics, don't use the mode that removes them
[a href="index.php?act=findpost&pid=355397"][{POST_SNAPBACK}][/a]

If I understood the OP correctly, the single guitar track was not from the same Soundgarden album, so album gain wouldn't make sense here.
Proverb for Paranoids: "If they can get you asking the wrong questions, they don't have to worry about answers."
-T. Pynchon (Gravity's Rainbow)

Replay gain is not perfect

Reply #17
It could also just be that his speakers/headphone's frequency response is out of whack in the treble. Which I'd guess is true for 90% of the listening environments out there actually.

Replay gain is not perfect

Reply #18
Using album gain does not solve the problem as I am comparing two tracks from different sources.

Simply put: if a human ear is able to adjust the volumes on two different tracks so that they both sound the same as for the loudness, so could the Replay Gain algorithm

It is true that poor mastering does a lot and that solo guitar is really filled with very high frequencies.

But still, I am able to adjust the volume manually so that it sounds more quiet and does not overshout Soundgarden. But it could be done by Replay Gain, couldn't it?

Replay Gain simply needs more investigation as for the psychoacoustics so that it can better decide between the real loudness among different frequencies. It should really act like a human ear and this is not an easy task.


note: I use Ogg Vorbis encoder with Vorbis Gain and Foobar 2000



Replay gain is not perfect

Reply #21
Quote
Quote
Have you even read the replaygain webpage?
http://replaygain.org/

EDIT: Spcifically this page:
http://replaygain.hydrogenaudio.org/calculating_rg.html
[a href="index.php?act=findpost&pid=356051"][{POST_SNAPBACK}][/a]


What does it solve? It will still sound the same.
[a href="index.php?act=findpost&pid=356053"][{POST_SNAPBACK}][/a]


It's plain rude that you complain about RG when it doesnt seem like you know anything about how it works and you act like you know what you are talking about when you say things like this:

Quote
Replay Gain simply needs more investigation as for the psychoacoustics so that it can better decide between the real loudness among different frequencies. It should really act like a human ear and this is not an easy task.


The RG algorithm does allready equal loudness filters, which you are suggesting it doesnt.
If manually adjusting RG values works better for you then keep usuing that method.

If you'r still not happy consider seing a psychiatrist

Replay gain is not perfect

Reply #22
Quote
The RG algorithm does allready equal loudness filters, which you are suggesting it doesnt.
If manually adjusting RG values works better for you then keep usuing that method.

If you'r still not happy consider seing a psychiatrist
[a href="index.php?act=findpost&pid=356058"][{POST_SNAPBACK}][/a]


I will try to post the files here in the future so that you can see that its method is not perfect.

Replay gain is not perfect

Reply #23
No one ever said it was perfect.

Replay gain is not perfect

Reply #24
Quote
What I mean, is the Replay Gain behaves too strictly or better to say it is too limited, it is not clever enough to analyze the real loudness in the tracks.

Because, as a result when I encode a Soundgarden track, and then a track where a single guitar player plays a standalone guitar with some high pass frequencies effects - the standalone guitar sounds too loud. And a human ear can clearly recognize that the standalone guitar is really louder then Soundgarden. As opposed, human ear would expect Soundgarden as a bunch of loudy guitars to sound louder...
[a href="index.php?act=findpost&pid=355247"][{POST_SNAPBACK}][/a]


OK, what's happening is that replaygain is making the two sounds play back at the same volume, but you THINK the standalone guitar is louder because you would naturally expect loud rock music to be played back louder than the guitar. So when replaygain plays them back at equal loudness, the guitar SEEMS louder.

Quote
the algorithm should put the volume level so that the frequencies appear more loudly in Soundgarden track I am pretty sure - in other words: it should recognize the critical frequencies that make the real loudness and put these frequencies at the same volume level that would make the human ear think that the traks are pretty much the same volume level even when thea are completely different as for the source of sounds

I think the Replay Gain does not corretly analyze the loud frequencies that human ear recoginzes as real loudness - and that is why it gives you the results that sometimes especially standalone instruments or voices are too loud compared to mixed instrumental sound.[a href="index.php?act=findpost&pid=355247"][{POST_SNAPBACK}][/a]


I understand what you're saying, but replaygain is doing exactly what it should. If I'm playing back using track gain, I WANT the guitar and the loud rock mix to play back at the same volume. I don't want replaygain making some decision regarding what plays back louder than what.

That said, I haven't heard your examples. Maybe you came across something that really throws off RG.