what is the best method of normalizing music (folk, metal etc.)

[geextah_2]

Hi, I try to normalize audio/music (big playlist with many genres, from Nu-Metal to Classic/Folk) I have already tried loudnorm and dynaudnorm. Dynaudnorm is very good but there are still big differences between tracks like 6-8dB, as it does not work with LUFS (human perceived loudness) but with dB/RMS, so its not possible to balance a (high compressed) Heavy Metal track to a (low compressed) Country/Folk song.
Loudnorm in 2pass can achieve that but I have no (real) experience using it. It also has (adds) more treble (sounds brighter) as it weights this frequency more (K-weighting filter) so it's kinda also an "Equalizer" which I don't know how to feel about.

Any ideas, tips?

Thanks for any help

[hello_hello]

I've only ever used ReplayGain myself. If a player supports reading the tags, it should adjust the tracks to the same volume without having to re-encode and/or compress them. If not, you can convert the tracks to the same volume using the ReplayGain info after scanning them. Have you tried ReplayGain?

foobar2000 can scan tracks using the EBU R128 scanning method and save the over-all volume in tags. It works the same way as the ReplayGain spec, only the EBU R128 scanning method is a little more accurate than the original ReplayGain algorithm. By default (I think) foobar2000 will adjust the volume on playback according to the volume info in the tags after you've scanned the tracks and saved the tags (an entire track is simply adjusted up or down in volume), so you could scan a bunch of tracks, save the ReplayGain data, and see what you think of the playback result before you convert and/or compress the audio itself, assuming you need to. The ReplayGain data can also be saved in SoundCheck tags for Apple devices to read, and foobar2000 can also use the ReplayGain info to physically adjust the volume of MP3 and AAC audio without re-encoding, or it can use the volume info to adjust the volume while converting. The default ReplayGain volume is -18 dB or -18 LUFS, whereas its -23 LUFS (I think) for the official EBU R128 standard and -16 LUFS for SoundCheck.

It's not the latest version, but I uploaded a portable version of foobar2000 a here while back. It has playlist columns for displaying the track volume after a ReplayGain scan (foobar2000 doesn't have them by default, and they're not necessary for ReplayGain to work, but I like to be able to see the volume info easily). It looks something like the screenshot below (the tracks being displayed below were physically adjusted to roughly -18 LUFS, hence the volumes being displayed are all close to that).

https://wiki.hydrogenaudio.org/index.php?title=ReplayGain
https://wiki.hydrogenaudio.org/index.php?title=ReplayGain#foobar2000_ReplayGain_scanner



PS. If you still want to compress the tracks (although hopefully you wont need to), you can compress them with Dynaudnorm, scan the compressed versions, save the volume info, then re-encode them while adjusting them to the same volume. The version of foobar2000 I linked to should have converter presets for compressing with Dynaudnorm, but unfortunately it's a 2 step process if you want to adjust the tracks to the same volume after they've been compressed.

[pandy]

my 2 cents - you can divide level adjustment (dynamic compression, loudness, normalization etc) on two ways, first non lossy as proposed by hello hello - replaygain or similar perform statistical analysis and modify your file by adding additional metadata but altering even single sample thus is lossless, second method assume altering PCM sample values i.e. performing re-quantization - unless your source is very high bitdepth (24 bits or similar) then this is lossy (by definition every requantization is lossy but with sufficiently high bitdepth and proper quantization error processing it can be considered as quasi lossless). By requantization you altering file permanently.

It is hard to advice on changing signal dynamics (this is how i understand your problem) - normally i don't like altering dynamics unless special purpose (like listening in noisy environment).
If you are strongly convinced about loudness normalization then perhaps https://github.com/slhck/ffmpeg-normalize can solve your problem.

[geextah_2]

From my understanding this https://github.com/slhck/ffmpeg-normalize
is just 2pass loudnorm which doesn't work for me, in the sense that I don't like how it sounds.

It doesn't matter if the change is destructive or non destructive, I keep the orignal files and this is only a copy on my phone.
One static volume adjustment doesn't work for me, there are loud parts and quiet parts in a song, so you have to dynamically adjust for it.

[geextah_2]

https://www.reddit.com/r/ffmpeg/comments/1mal7jx/perfect_music_normalization_with_dynaudnorm_lufs/
I made my own script for that

[Barrythecrab]

Levelator

[geextah_2]

Levelator looks nice, but it does not support command line and also only works with wav, it is also meant for podcasts and uses RMS instead of LUFS

[Barrythecrab]

Agreed Levelator is really sparse in its offerings! Bare is being generous, but I thought I’d toss into the fray.

[geextah_2]

yeah sure on paper it sounds nice, limiter compressor normalizer, all of these things are needed, but it would need command line for batch conversion, what will you do with 30.000 files, drag and drop them one by one?
you can see that it was build with podcast in mind, so you have one giant wave file and then it makes sense for that purpose

I am still working on my script (dynaudnorm + LUFS) and its actually "finished" it uses ffmpeg so you can do whatever you want with it. It hits ~99% a LUFS range of 13,3-15,2 (Average ~14,2 LUFS) It uses 3 passes, the first one does ~90% of the job, then the 2nd and 3rd pass are for the ouliers/rogue files that need more adjustment.
There will be 2 versions, one that will keep your subfolder structure, if you have a bigger library and one that will just ouput all files in one directory.

[DOMNG]

For my part, I ended up preferring loudnorm in 2-pass, it is the one that gives the most coherent rendering between very different styles. Yes, it colors a little, but that's the price to pay for a real homogeneity of volume.

Tutuapp

[geextah_2]

For most people thats probably fine. I also wanted to use it but then I started testing it and in some files the silent part remained silent and the loud part was clipping.
I updated my script on reddit, you can try it out if you want.
I run it on about 30k files for my kitchen radio, and so far it sounds balanced well.

https://www.reddit.com/r/ffmpeg/comments/1mal7jx/perfect_music_normalization_with_dynaudnorm_lufs/

[Turp]

Hi guys, this thread is interesting and I think comes closest to my question (hence replying here rather than creating a new thread).

Let me start by admitting that most of your previous discussion is beyond my current understanding, however I have recently learned the difference between peak and lufs normalization, and also the basics of what a limiter and compressor do.


There's a YouTube channel (https://www.youtube.com/channel/UCN8ewC6LFb0YnqHpw9XsN5Q) with about 60 songs which are fairly close to the same genre (bluegrass southern revival), and I'm pretty sure that it's all AI generated with various pro tools (singing is absolutely pitch perfect and the instruments especially the banjo sound like an old cheap Yamaha keyboard sample). 1. some albums/releases (10-20 songs) are at a pretty different loudness level, and 2. at least one song in particular has several pretty drastic volume changes at at least 3 places (which seem very intentional rather than YouTube doing it). Here's the song that I'm talking about: https://www.youtube.com/watch?v=x0bgytzREak

So my question is how can I level out not only all of the songs with each other, but also more importantly scan and fix any songs that were obviously mastered badly/incorrectly? I can use audacity's peak and loudness normalization effects to manually correct it, but that's a lot of work.

[hello_hello]

So my question is how can I level out not only all of the songs with each other, but also more importantly scan and fix any songs that were obviously mastered badly/incorrectly? I can use audacity's peak and loudness normalization effects to manually correct it, but that's a lot of work.

Try foobar2000.
If the songs are okay individually, ReplayGain is the way to go.

foobar2000 portable (for audio encoding)

Load a bunch of files into a playlist (it's okay if they contain video), select them all, right click and select "ReplayGain/Scan per file TrackGain". When it's done a new window will open. Save the volume info to the audio files.


[Attachment 92186 - Click to enlarge]

The tags will contain the volume adjustment for each track. In a perfect world a player would use that info to adjust the volume on playback and that'd be it. In the real world though, it's probably better to adjust the audio volume rather than rely on a player reading the tags. For MP3 and AAC audio you can right click and select "ReplayGain/Apply Gain to file content". That'll physically adjust the volume in each audio frame. A new window should pop up first. For the moment only select:

Apply track ReplayGain
Make files louder or quieter
and leave the volume slider on 89dB.


[Attachment 92187 - Click to enlarge]

When it's done the track gain info should change so it's very close to 0dB, which means the audio is at the target volume. A small variation is normal as to adjust the volume without re-encoding it must be adjusted in increments of 1.5dB. Formats other than MP3 and AAC need to be re-encoded to adjust their volume (you can configure/add ReplayGain to the converter).

For the record, ReplayGain doesn't compress or change the dynamic range. It simply determines the average volume of each track and whole tracks are adjusted up or down.

For the problem tracks, try converting the audio using one of the compression presets. My go-to compression is labelled:
Converter / Compression - Dynamic Audio Normalizer, Single Thread
but try the conversion preset below it labelled (possibly only supports stereo):
Converter / Compression - LoudMax, Single Thread.
The audio tracks will be compressed on their way to being re-encoded as AAC.

Assuming compressing the audio does even it out, you'll need to run ReplayGain again on the re-encoded files. While each output file will hopefully have a consistent volume, they probably won't have the same average volume as the non-compressed files, so scan the new files, save the ReplayGain info as before, and if the output was MP3 or AAC you can adjust the volume with "ReplayGain/Apply Gain to file content".


[Attachment 92188 - Click to enlarge]

A note on ReplayGain's retarded way of specifying volume.
The ReplayGain target volume of 89dB is based on a SMPTE standard for sound pressure level in theaters, and as such it's meaningless to most mortals, however....

ReplayGain's 89dB equals a volume of -18dB on an audio player's output meter, where 0dB is the maximum.
The ReplayGain volume data saved in tags is always relative to -18dB (-18LUFS), so when you see a track gain of +3.5dB, for example, it means the volume is -21.5 dB.
Some programs let you change the target volume, although the default is sensible and shouldn't be changed without good reason (standard music tracks), but if you adjust the volume to something other than the default, the tags are still written relative to -18dB.

Depending on where you live, the standard volume for broadcast TV is roughly -23dB (-23LUFS). To adjust to -23dB you'd need to change the ReplayGain target volume to 84dB.

Here's what the volume info for the sample you linked to looked like.
1 - Original audio, adjusted to -18dB
2 - Audio compressed and re-encoded. The ReplayGain scan shows the volume increased.
3 - A copy of the compressed audio after adjusting the volume again with ReplayGain.

Sorry if the compression didn't have the desired affect, but it's so noisy where I am at the moment I didn't try to listen to them.


[Attachment 92185 - Click to enlarge]

I didn't plan to make this post such an essay, but as the version of FB2K I linked to is a little old, I'll upload the latest stable version with my configuration tonight.

[Alwyn]

I second Foobar 2000. I use it to volume-balance my MP3s and all my music video MP4s. It works very well. I never have to grab the volume knob to adjust it.

[Turp]

So my question is how can I level out not only all of the songs with each other, but also more importantly scan and fix any songs that were obviously mastered badly/incorrectly? I can use audacity's peak and loudness normalization effects to manually correct it, but that's a lot of work.

Try foobar2000.
If the songs are okay individually, ReplayGain is the way to go.

I didn't plan to make this post such an essay, but as the version of FB2K I linked to is a little old, I'll upload the latest stable version with my configuration tonight.

Wow, thanks very much for the details. I downloaded the latest stable Foobar2000 and will mess around with some test files and settings to better understand what and perhaps how it does what it does.

It also has DSP capabilities which is very cool! I've used a very old version of OTSDJ for many years which usually makes most music sound better (more "alive" like radio music). It also does 5 sec auto mixing between tracks.

I used Audacity to manually fix the volume using regular normalization because I could actually see where the three main volume changes occurred. Then I ran all 60 of their songs through MP3Gain using both track gain first and then album gain. That got me close, but I wanted something more accurate, consistent and perhaps easier.

I listened to all three files and the volume changes are still present in all three. It's very weird, like they manually dropped the volume knob a few notches in several places. And it seems like it's only on that one song, although their Zydeco release and newer songs are about 20% louder than their previous stuff. The noticeable volume changes in this sample song are at:

0:29 down significantly
1:30 up slightly
2:58 down slightly
3:27 down significantly
3:57 down significantly

I use Plexamp to stream my library which works ok, but I'll see if FB2K is better.

[hello_hello]

Just so you know, there's a newer standard for determining volume with the fun name of EBU R128. It's better than the older Replay Gain algorithm. Even though the scanning function is labelled Replay Gain in foobar2000, it actually uses the EBU R128 method by default (although it can be changed in preferences) so it's scan results won't be exactly the same as MP3Gain.

Foobar2000 also doesn't have the ability to undo volume changes. MP3Gain can undo the changes because it writes the information it needs to it's own fields when it updates the tags, although if the tags are over-written by another program, the undo information can be lost.

Mp3Gain saves the ReplayGain info to Ape tags, the idea being that most programs use ID3v2 tags, so generally they won't mess with the Ape tags. Foobar2000 can read MP3Gain's tags and I think it deletes them if it updates the ReplayGain info, but MP3Gain is oblivious to ID3 tags so if you save ReplayGain info with both programs and change the volume you can end up with conflicting info. If you've scanned files with one program it's probably a good idea to use it to delete that info before you intend to scan and adjust the volume with the other.

Foobar2000 has a compressor DSP called EBU R128 Normaliser. It uses the algorithm to apply compression and keep the volume at -18 dB (I think). You could try that, but there's a few different compression DSPs and most of them are configurable.

Edit: I've update the thread for the portable version of foobar2000 I linked to now. The opening post now contains a link for version 2.25.8
foobar2000 portable (for audio encoding)

I still use the 32 bit version of FB2k because one of the GUI components (Facets) had to be re-written from scratch by the author of FB2K for a 64 bit version, as whoever created the original component is MIA, but it doesn't have all the features of the 32 bit version. I also use the Text Display component in the GUI and it only comes in a 32 bit flavour.

PS. I'm fairly sure that by default, FB2K will read any ReplayGain info in tags and adjust the volume on playback, so once it's saved the data, all the tracks should play back with the same volume. If you still want to hear the volume difference after you've scanned your files and saved the ReplayGain info, you need to disable the ReplayGain adjustment on playback in preferences.

[timtape]

...It is hard to advice on changing signal dynamics (this is how i understand your problem) - normally i don't like altering dynamics unless special purpose (like listening in noisy environment)...

Yes there's the view out there that any track or playlist of tracks can be "mastered" to sound great no matter what the playback system or the other listening conditions. It's a pipe dream. The same tracks played in a very noisy and then a very quiet environment will sound quite different and need different treatment. For one, we're often up against the limits of human hearing and the masking effect. There's no magic wand or one size fits all.

[lordsmurf]

I only entered this thread because it had "metal" in the title.

Where are the metalheads at?