Volume Level Equalizing a Bunch of MP3 files. How to do it?

[cornemuse]

I have a bunch (1,000's) of mp3 files, the volume levels are mostly ok, but some are pretty low volume.

I have Audacity for mainly converting vinyl to mp3/cd. It also works fine for increasing volume on single files.

Its been my experience that batch processing, (with those progs I've tried) they tend to balance by lowering volume to match the lowest level file.

Is there a batch type of program that will make 'em all same volume level at a substantially higher level??

I am very hard of hearing, and need all the help I can get!


-corne-

[sneaker]

http://mp3gain.sourceforge.net/

Set all to the same high volume you want. Enjoy your clipping.

[hech54]

First you need to pick "the one" that all of the other must follow/change to.

[cornemuse]

First you need to pick "the one" that all of the other must follow/change to.

That would be the loudest one!

Now, I cant tell, but does this process degrade the quality of the music? Some I have increased the volume by a major amount and they still sound just fine, just wondering, , , thats all

-c-

[pandy]

you can use dynaudnorm filter (from LoRd_MuldeR http://muldersoft.com/docs/dyauno_readme.html ) - it should be integrated in SoX and it is already integrated in ffmpeg. For example for my personal usage i prefer dynaudnorm with peak at -3.103dBFS and mild dynamics compression - this filter is safe way to perform automatic gain and partially also loudness alignment on many files (bellow example for ffmpeg).

-af dynaudnorm=p=1/sqrt(2):m=100:s=12

As ffmpeg use libmp3lame then i believe quality can be also more than decent...

@ffmpeg.exe -y -hide_banner -loglevel 24 -stats -i "%1" -vn -c:a libmp3lame -q:a 1 -af dynaudnorm=p=1/sqrt(2):m=100:s=12 "%~n1_.mp3"

[Hoser Rob]

... I cant tell, but does this process degrade the quality of the music? ...

Well, yes, it will definitely decrease quality. When I'm playing media on my laptop I use a Beringer USB DAC and just tell the playback software to send the audio stream straight to that device, no resampling or conversion. This is easy in L:inux.

But if you can't tell ... and are you really sure about that? ... it won't make much difference.

But that assumes that normalization isn't causing signal clipping, as mentioned. You'll definitely notice that. It sounds horrible.

You also cannot apply global normalization and expect all the files to play at exactly the same volume. It depends on the levels in the input file. Ironically if you normalize a highly compressed recording (like most all popular music in the last 15 years or so) it'll play less loudly than a recording using the same normalization which asn't compressed so much, which is the norm with older recordings.

I'm wondering how you're playing these files. Any computer software player I know of has built in normalization. I'm assuming that because I do normalize movie audio but never with music, but I think that's a safe assumption.

[sneaker]

Now, I cant tell, but does this process degrade the quality of the music?

Unlike most other methods using mp3gain does not affect the quality of the files. But it is limited in that it only increases/decreases volume by the same amount in all of the file. It cannot do things like you can do with dynaudnorm mentioned by pandy. So like Hoser Rob mentioned maybe a solution on the player level would be more adequate and spare you the work of having to mess with the files themselves.

[pandy]

Subjectively dynaudnorm will not affect audio quality... mp3 is not audiophile file format - there are millions of the mp3 codec consumers where audio quality degradation related to using dynaudnorm will be unperceived at all...

[sneaker]

dynaudnorm is only the first step of degradation. If you want to have mp3 as output you have another step of degradation: lossy re-encoding. But I agree that subjectively people will probably not be able to tell. But maybe in a year he will find another even better suited normalization software. Then he might degrade again...
But that's just food for thought.

[pandy]

dynaudnorm is only the first step of degradation. If you want to have mp3 as output you have another step of degradation: lossy re-encoding. But I agree that subjectively people will probably not be able to tell. But maybe in a year he will find another even better suited normalization software. Then he might degrade again...
But that's just food for thought.

Well problem is in mp3 itself - however it is proven as high quality audio codec for most of us...

[cornemuse]

... I cant tell, but does this process degrade the quality of the music? ...

Well, yes, it will definitely decrease quality. When I'm playing media on my laptop I use a Beringer USB DAC and just tell the playback software to send the audio stream straight to that device, no resampling or conversion. This is easy in L:inux.

But if you can't tell ... and are you really sure about that? ... it won't make much difference.

But that assumes that normalization isn't causing signal clipping, as mentioned. You'll definitely notice that. It sounds horrible.

You also cannot apply global normalization and expect all the files to play at exactly the same volume. It depends on the levels in the input file. Ironically if you normalize a highly compressed recording (like most all popular music in the last 15 years or so) it'll play less loudly than a recording using the same normalization which asn't compressed so much, which is the norm with older recordings.

I'm wondering how you're playing these files. Any computer software player I know of has built in normalization. I'm assuming that because I do normalize movie audio but never with music, but I think that's a safe assumption.

MP3's on an old Argosy media player, audio feeds to older stereo amp, (how old? has phono input & no remote!). vid to otherwise worthless 6" Haier portable TV, so I can see where/what I'm playing, 200 +- files each in about 12 folders. Turn tv off once music starts. Stereo out to Russound speaker selecter to various rooms in the house.

I worked on/around jet planes (usaf) then 40 years around heavy equipment, yes, this did a major number/hit on my hearing, I'm sure, I definitly cant hear very good.

By 9:00 am, the computer is off for the rest of the day usually, only listen on comp when ripping, editing, etc vinyl/cd. Argosy has usb input, easy to add/test music, , ,

-c-

(Typically, I dont watch movies without subtitles, need that to follow the film, , , )

[cornemuse]

I copied a batch of mp3s to a diff folder (so if I screwed things up, it was copies) & tried 'mp3gain', worked ok, info for usage seems a bit lite, tho. Just 'clicked' through, satisfied with the results, , , ,

Thanks, , , -c-

[hello_hello]

I would have suggested mp3gain too. Scan all MP3s using Track Analysis, then apply Track Gain to adjust the volume. They're all adjusted to the default ReplayGain target volume of 89dB, which is somewhat meaningless without an explanation, but the general idea is to not change it and always adjust to the same volume.

Foobar2000 can scan MP3s loaded into a playlist. It uses a newer algorithm than the original ReplayGain algorithm which is considered to be more accurate. You'd load the MP3's into a playlist, highlight them all, right click and select "ReplayGain/Scan per file Track Gain". When it's done, save the info to the files, right click again and select "Apply Track Gain to File Content". It'll adjust to the same 89dB target volume.

Mp3Gain is probably a bit easier to use in that it's GUI makes individual file volumes and peak levels easier to see. It saves "undo" information to tags so the process is reversible (as long as the tags aren't removed).
Foobar2000's scanner probably a bit more accurate, it's about 100x faster when scanning a single file and it scans multiple files at a time. Foobar2000 doesn't save undo information. It can also scan and losslessly adjust the volume of AAC in an MP4 or M4A the same way (Mp3Gain can do the same but requires a separate download and replacing one of it's exe files). Foobar2000 also has options for re-writing MP3 headers and rebuilding MP3s from scratch, which can be useful for fixing problematic MP3s. You can change the 89dB target volume, but Mp3Gain makes that easier to do. For foobar2000 the setting is buried deep in it's preferences.

[V1de0Luvr]

you can use dynaudnorm filter. For example for my personal usage i prefer dynaudnorm with peak at -3.103dBFS and mild dynamics compression (bellow example for ffmpeg).

-af dynaudnorm=p=1/sqrt(2):m=100:s=12

Thanks for the tip. I am trying to use this in FFmpeg but can't figure out your options? I want -3db from 100% gain, which may be close to the .95 default peak. I have been using WaveGain which defaults to -3db but only supports 2-channel WAV files, and I want to normalize 5.1 channel AC3.

Strange, going form AC3 to WAV, -af dynaudnorm=p=0.35 gives me what sounds like the same volume level as using WaveGain after downmixing AC3 5.1 to stereo.

[pandy]

Thanks for the tip. I am trying to use this in FFmpeg but can't figure out your options? I want -3db from 100% gain, which may be close to the .95 default peak. I have been using WaveGain which defaults to -3db but only supports 2-channel WAV files, and I want to normalize 5.1 channel AC3.

Strange, going form AC3 to WAV, -af dynaudnorm=p=0.35 gives me what sounds like the same volume level as using WaveGain after downmixing AC3 5.1 to stereo.

Peak in my case is already -3.103dBFS (solve 1:SQRT(2)) and convert to dBFS - it will be exactly -3.103dBFS). My settings introduce two differences from default dynaudnorm - i allowed higher level change (100) this is to deal with extremely silent sources where up to 40dB "amplification" can be applied and second kind of mild dynamics (traditional) compression to further normalize loudness).

[V1de0Luvr]

Peak in my case is already -3.103dBFS (solve 1:SQRT(2)) and convert to dBFS - it will be exactly -3.103dBFS). My settings introduce two differences from default dynaudnorm - i allowed higher level change (100) this is to deal with extremely silent sources where up to 40dB "amplification" can be applied and second kind of mild dynamics (traditional) compression to further normalize loudness).

Thanks, I couldn't find any documentation about using a mathematical equation for the P setting. I had to give up the idea of normalizing 5.1 music since the 0.35 setting played way too soft on a 5.1 system. Playing 5.1 audio on a stereo system sounds louder, so I am going back to a stereo mixdown with Dolby Pro Logic II encoding to ensure that all of my music videos sound good together.

After the stereo mixdown I found that dynaudnorm=p=0.70 gave me a similar level as WaveGain at -3db. The m=100 did not seem necessary for music, but I'll use it for movies where I am keeping the 5.1 audio.

[pandy]

Thanks, I couldn't find any documentation about using a mathematical equation for the P setting. I had to give up the idea of normalizing 5.1 music since the 0.35 setting played way too soft on a 5.1 system. Playing 5.1 audio on a stereo system sounds louder, so I am going back to a stereo mixdown with Dolby Pro Logic II encoding to ensure that all of my music videos sound good together.

After the stereo mixdown I found that dynaudnorm=p=0.70 gave me a similar level as WaveGain at -3db. The m=100 did not seem necessary for music, but I'll use it for movies where I am keeping the 5.1 audio.

Peak equation is simple - Sine(45deg) or 1/sqrt(2) - in both cases you have exactly half bit level reduction or -3.103dBFS - 0.7 is just approximation for 0.707....

[V1de0Luvr]

Peak equation is simple - Sine(45deg) or 1/sqrt(2) - in both cases you have exactly half bit level reduction or -3.103dBFS - 0.7 is just approximation for 0.707....

FYI, I did some testing between your method and a simple 0.70 and can't tell the difference.

BTW, I thought I might like the automatic fade-in and fade-out, but had to add b=1 to disable that.

[pandy]

Peak equation is simple - Sine(45deg) or 1/sqrt(2) - in both cases you have exactly half bit level reduction or -3.103dBFS - 0.7 is just approximation for 0.707....

FYI, I did some testing between your method and a simple 0.70 and can't tell the difference.

BTW, I thought I might like the automatic fade-in and fade-out, but had to add b=1 to disable that.

Well... it is not my method - this is simple math relation - 0.7 is slightly less optimal in terms of using hypothetical dynamic range but difference will be marginal especially on 16 bit or higher bit depth audio - but if you trying to squeeze every fraction of a bit this can be important, also it may marginally increase quantization noise but i assume barely measurable and not perceivable.

[V1de0Luvr]

Well... it is not my method - this is simple math relation - 0.7 is slightly less optimal in terms of using hypothetical dynamic range but difference will be marginal especially on 16 bit or higher bit depth audio - but if you trying to squeeze every fraction of a bit this can be important, also it may marginally increase quantization noise but i assume barely measurable and not perceivable.

Can't say I understand any of that but I do thank you for pointing out the ffmpeg filter. After further testing I settled on:

-af dynaudnorm=p=0.70:m=100:g=11:s=12

With b=1 I sometimes saw the start & end amplified too much. Reducing Gauss to 11 makes fade-in/out work much better, and compression was almost unnecessary, but still helps a little.