normalise: automatic or bespoke

[hooty2]

Hi...
I've been tasked to edit some recordings of a series of improvised cello & harp performances (simply cut dialogues).
I was then asked to normalise the files (MP3)
I watched reaper videos (Kenny Gioia...whom i do like), seemingly treat each item/track etc, by hand according to need with tips to ensure the original dynamic range/level ratios are preserved etc. (ie: caution with global settings)
Is this the way to go?
The tracks typically peak around -4dB on crescendos, with quitest material around -20dB, median range around -18dB/-14dB
what do people say?

[hooty2]

p.s.
any comments/guidance on rendering with constant bit rate or maximun bit rate/quality?
cheers

[ken long]

seemingly treat each item/track etc, by hand according to need with tips to ensure the original dynamic range/level ratios are preserved etc. (ie: caution with global settings)

Seems sensible for discrete recordings.

If they are all related and part of a single performance though, you may want to create a contiguous file out of them and normalise the whole thing so that you get consistent processing throughout.

I usually normalise to -1dBFS peak value.

[hooty2]

Thanks for that Ken.... i'll be finished now before breakfast

[Mike Stranks]

seemingly treat each item/track etc, by hand according to need with tips to ensure the original dynamic range/level ratios are preserved etc. (ie: caution with global settings)

Seems sensible for discrete recordings.

If they are all related and part of a single performance though, you may want to create a contiguous file out of them and normalise the whole thing so that you get consistent processing throughout.

I usually normalise to -1dBFS peak value.

I agree in all particulars!

[hooty2]

[Hugh Robjohns]

I was then asked to normalise the files (MP3)

Do you mean you're normalising WAV files to release as .mp3s, or are you working on .mp3 source files?

Processing .mp3 files -- especially changing level or EQ -- isn't a great idea because it messes with the original coding assumptions and results in more audible artefacts.

I watched reaper videos (Kenny Gioia...whom i do like), seemingly treat each item/track etc, by hand according to need with tips to ensure the original dynamic range/level ratios are preserved etc. (ie: caution with global settings). Is this the way to go?

It very much depends on the material and the reason you're normalising the tracks. If it's an album of work, it would be foolish to normalise the most quiet and gentle tracks to the same peak level as the loudest and brashest! That would rob the album of its internal dynamics completely. On the other hand, if these are independent tracks that you want to all sound roughly the same, then normalising each separately would be a better approach.

That said, peak normalisation is now an outmoded concept and you really should be working with loudness normalisation (aiming for -14LUFS or -16LUFS, or whatever the target loudness figure is for your intended destination. With loudness normalisation, the only concern over peak levels is that they remain somewhere below -1dBTP.

The tracks typically peak around -4dB on crescendos, with quitest material around -20dB, median range around -18dB/-14dB
what do people say?

Normalising only changes the peak level. It does nothing to change the dynamic range. If you need to alter the dynamic range -- the ratio of peaks to quiet or median bits -- you'll need to think about limiting, compressing, parallel compressing, etc.

A peak level is -4dBFS isn't bad and personally, I probably wouldn't bother to squeeze another couple of dB out of it. Few would notice the difference, and if this is material to be encoded as .mp3 you will need some headroom for the encoder anyway.

For pure .wav material many will advocate normalising to -1dBFS. However, all normalising algorithms I know work on sample amplitude values and don't take inter-sample peaks into account at all. So I would recommend checking with a true peak meter after normalising to make sure your processing hasn't inadvertantly created inter-sample peaks.

If processing for any lossy codec I'd also do some experiments to find the optimum peak level as it's quite common for the codec to run out of headroom during the file conversion. The EBU recommend peaks no higher than -3dBFS for material destined for lossy codecs and I've found that to be good advice... but different codecs (even of the same output format) have different requirements. (In the case of .mp3, the spec only defines the decoder algorithms, and so different encoders can and do work in different ways!)

[The Elf]

I was going to raise a similar point about MP3s - this should be a FINAL format and definitely should not be worked on once created.

Unfortunately I often get work from people who think that MP3 is just another audio format, like WAV, or AIFF, not realising that a lot of the data has been thrown away by the encoding process.

So I'm re-iterating Hugh's point to emphasise it here. We really need to consign MP3s (I'd include all lossy formats) to history!

[Aaron Straley]

What if you were given mp3 tracks to work with and have no choice?

Would you treat these tracks differently in any way

[Hugh Robjohns]

What if you were given mp3 tracks to work with and have no choice?

Would you treat these tracks differently in any way

I would mutter and grumble while sneering as I worked on them!

Slightly more seriously, I would do everything possible to get better quality source files to work with, and try to re-educate the person supplying the files as to what they should be supplying and why.

From a practical point of view, there's nothing you can do to replace the missing audio content, but you can synthesize some new material to assist any future lossy codecs and try and minimise the concatenation errors. One method I've used is to mix in some subtle 'ambience' or room 'reverb', or even some low level broadband noise, depending on the material.

If you listen to DAB radio much, it's surprisingly common to hear the effects of lossy codec concatenation when someone in the broadcast studio plays an .mp3 file or some other lossy-codec material ripped off the interweb (audio from streaming TV clips is common). This material almost certainly sounded absolutely fine in the studio, but it comes out sounding very mangled after a second pass through the lossy codec in the DAB transmission chain!

[The Elf]

What if you were given mp3 tracks to work with and have no choice?

Would you treat these tracks differently in any way

Sometimes this is the case.

Obviously it's not ideal, but as a professional I just get on with it, though I will try to educate, and I will, by necessity, point out the compromises.

Buried in a mix you can get usually away with it, but anything exposed can often benefit from a little excitation or some filtered distortion (and reverb can hide a multitude of sins), depending on the nature of the audio.

[Aaron Straley]

All the mp3 tracks are harmony vocals recorded mainly on cell phones. I am doing this for my own education and also as a favor for a friend. This is a good project for me right now, as I don't have much experience mixing harmony vocals. The original mix job for the harmony vocals was absolutely butchered, and I am certain I can improve it.
I wont hijack this thread, but once I get further along, will post on here for you all to review.

[James Perrett]

That said, peak normalisation is now an outmoded concept and you really should be working with loudness normalisation (aiming for -14LUFS or -16LUFS, or whatever the target loudness figure is for your intended destination. With loudness normalisation, the only concern over peak levels is that they remain somewhere below -1dBTP.

The OP mentioned Reaper. In Reaper it is easy to do loudness normalisation if you have the SWS extensions installed. Open the Extensions->Loudness dialog, click the Options button and make sure that you have the Measure True Peak option selected. Then select the items that you want to normalise in Reaper's main window, choose "Analyze selected items" in the loudness window and you will see the loudness statistics for each item. You can then right click, choose normalize and then normalise to whatever level you choose.

Normalize simply tells Reaper how loud to play each item, if there is an odd peak over your chosen peak level then you can use ReaLimit in True Peak mode on the track (or master) to limit that peak.

However, I would add that I never accept Reaper's default normalisation without listening and I will often tweak the levels by a dB or two until it sounds right to me.

[The Elf]

All the mp3 tracks are harmony vocals recorded mainly on cell phones.

It sounds like they're probably going to be pretty ropey anyway, so all bets are off. In those situations you do what you can.

There's one recently released track where I had to artificially create vocal 's' and 't' sounds with filtered synth noise, for a part recorded on a phone, but that's another story...

[Mike Stranks]

These days I get tracks in all sorts of formats...

Standard practice now is to convert immediately to 32-bit FP and leave them like that all the way through the editing, cleaning, 'fairy-dust' processes.

They often have to end up as MP3s. That's the very last thing I do after everything else has been done. I'm effectively making an MP3 copy of the finalised recording.

[Hugh Robjohns]

Standard practice now is to convert immediately to 32-bit FP and leave them like that all the way through the editing, cleaning, 'fairy-dust' processes.

That would be standard (automatic) practice in most DAWs.

They often have to end up as MP3s. That's the very last thing I do after everything else has been done. I'm effectively making an MP3 copy of the finalised recording.

Always listen to the final .mp3 all the way through to check for concatenation errors. Don't assume it will be the same as the sound coming from your DAW main outputs. That extra lossy codec stage can do a lot of damage in some cases (and next to none at all in others!)

[Mike Stranks]

Standard practice now is to convert immediately to 32-bit FP and leave them like that all the way through the editing, cleaning, 'fairy-dust' processes.

That would be standard (automatic) practice in most DAWs.

Ah! But I'm not using a DAW for most of my work...

Standalone Audio Editor then RX7 in standalone mode... Yup; I know, but it's what I'm used to... DAW usually means 'thinking time... my way is semi-auto, 'cos I've been doing it so long...

[Hugh Robjohns]

Same applies. It's virtually impossible to manipulate .mp3 files natively, so whatever 'standalone editor', audio restoration system or DAW you use they will all decode the .mp3 to the system's native internal DSP format which is commonly 32 bit floating point or 64 bit floating point, depending on your computer OS.

[hopscotch44]

.... excellent information and discussion. Thank you.
I received the files as mp3.
I only had to edit out dialogue & present a collection of music tracks as mp3.
Dropped into Reaper/44.1 - 24 bit. (was this wrong? )
The origin of the mp3's was a take from the console at a local festival.
My understanding is that these are for reference and archive.
The programme material is improvised violin and harp.
There is a wide dynamic range.
The only processing was on a small section of one track where an internal fx loop happened for approx 10 secs (organisers/composers asked for tons of reverb FOH.... i hear it was a yamaha desk, and i noted a previous thread of mine concerning fx loops on my 01v96) i split the track at that section & removed with RX. (i then basked in the reflected glory of RX/spectral repair)
I was asked to normalise... and i confess to never having done this and i was not sure how or why!

So i did this:
I put FF ProL2 limiter on each... preset - safe/transparent musical - KMeter -20. True peak & oversampling x4.
I added gain to most tracks to achieve peaks on crescendos of approx -4/-5dB.
Here and there the instruments did a pluck or something that was caught by the limiter.
I listened through all the material.... but not the final renders
I will do that today.
Personally i feel a little manual adjustment was needed but discreetly.
I have learnt from SOS in other threads, about manipulating files in varying formats (and i am unsure yet as how to best manage that scenario and i liked the references to that in the thread.... research to follow )
I'm extremely grateful for this response from everyone.
Cheers

[The Elf]

Ideally I would go back to the client and ask for WAV files, and if this is not possible then also suggest that future recordings are not done to MP3 directly! Most people are happy to change - they just don't realise what the implications are.

By starting with MP3s, processing then re-encoding again as MP3s, irreparable damage is being done to the audio - and will likely be audible.

A few years ago I had a band that copied their audio around by writing MP3s to CD, ripping to MP3, writing to CD... repeating the process for each band member. By the time the bassist had a copy the audio sounded like it was coming out of a tin can!

[Mike Stranks]

Same applies. It's virtually impossible to manipulate .mp3 files natively, so whatever 'standalone editor', audio restoration system or DAW you use they will all decode the .mp3 to the system's native internal DSP format which is commonly 32 bit floating point or 64 bit floating point, depending on your computer OS.

Of course...

... but I have to work with what I'm given... usually from techno-phobic, 'mature' people who need instructions of the 'you’ll now see this screen; click this button' variety and frequently still need a phone conversation of the 'you should now be able to see the runway ahead of you... ' variety. Just getting a recording from them to me is a major achievement.

I concur heartily with all that's been said about 'I don't want to be working with MP3s', but sometimes it's that - or other lossy formats - or nothing at all for me. In my context of (very) amateurishly recorded spoken word, then often the conversion from MP3 and back again is the least of my quality worries...

I think we're saying the same thing... I'm just pointing out my realities.

[Hugh Robjohns]

Realities fully understood, Mike. Worn out those t-shirts myself, too. Such compromises are, sadly, necessary sometimes but we should always strive for better and, in cases like the OPs, explain why it's important.

The thing to remember is that .mp3 along with most other lossy codes formats is/are designed for use as a convenient final delivery medium to the consumer.

Using them within the production/ post-production chain is playing with fire.