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.