WAV vs MP3

[stringbean01]

i'm mixing an album and am running into the strangest thing... ive never noticed this with any other music before. when i bounce my masters straight from logic to WAV, all transients are clean and quite beautiful sounding. but anytime i encode to MP3 (whether it be straight out of logic on the "highest" quality VBR 256, or from itunes) a lot of transients distort in playback.

i'm mastering with voxengo elephant as the final plugin in the chain. mix reads under 0db without question in logic and my loudness/small dynamic range isnt earth shattering... what is happening? what is the root of this problem! the mix is great and the EQ curve is quite standard and comparable to much of todays music. WAV sounds awesome. MP3s suck on really high quality settings and a "reputable" encoding algorithm

[lookingfor]

The highest quality setting for MP3 is 320 CBR (costant bit rate).
The quality of mp3 conversion depends also from the encoder.
Try other stuff, like Max (freeware).
With that bitrate you should not notice big differences.

[Marando]

When I encode my music to mp3 256kbit ABR with HQ setting, using WinLAME, I can't hear the difference in a blind test. However, this is when I make sure the original wav file maxes out at -0,5dB.

When the original wav file maxes out at -0,1dB, I allmost always have audio clipping in the mp3 file, even when I encode it in 320kbit.

Anyone else experienced the same?

[Jon Hodgson]

I don't know this for a fact, but sensibly one might expect iTunes and Logic to use the same mp3 encoder, since you'd hope that Apple would want their best in both.

So maybe try a completely different one, say LAME, to see if that makes any difference.

You say the EQ curve is fairly typical, have you looked at a spectral view? Stuff outside of the mp3 bandwidth shouldn't affect a properly implemented encoder, but we don't know if it's been properly implemented, try low pass filtering above 16kHz, and see if the encoder manages better.

Finally, why mp3? AAC is superior and I think you'll find it on most players these days.

[domokunrox]

Your problem is bouncing masters out of Logic.

You think their WAVs sound good? They make my ears want to commit suicide.

It has this weird, smeared, distorted low end characteristic. Might sound good for Miami booty bass records.

[stringbean01]

Originally Posted by domokunrox

Your problem is bouncing masters out of Logic.

You think their WAVs sound good? They make my ears want to commit suicide.

It has this weird, smeared, distorted low end characteristic. Might sound good for Miami booty bass records.

unfortunately not all of us have years and years of wise, multifaceted expertise on the subject. i come here to learn! certainly i feel trapped in the box... do you find PT to write better WAVs? i cant wait to own a console....

anyway, thanks. i'm gonna try a tad higher of a high pass. SOS/TOMB forums also suggested that i bring the peak down to -.03 or .05.

although, that leads me to a new question... most commercially distributed, professional records peak clean at 0 dont they? i feel like i see that all the time when loading reference tracks in logic. i shouldnt HAVE to sacrafice my mix peak to solve this. its gotta be a frequency content thing.

[stringbean01]

ok yeah im an IDIOT. this is a 24 bit project and i have no dither (didnt enable it in a plugin and have it check off in logic's bounce.

this can only help the situation.

[Alexey Lukin]

Originally Posted by Marando

When the original wav file maxes out at -0,1dB, I allmost always have audio clipping in the mp3 file, even when I encode it in 320kbit. Anyone else experienced the same?

Yes, this has been already discussed on this board. Any lossy encoding changes peak signal levels, every peak can randomly become higher or lower. This often leads to clipping, unless some headroom is left.

[It'sJoeAgain]

Originally Posted by aemaury

ok yeah im an IDIOT. this is a 24 bit project and i have no dither (didnt enable it in a plugin and have it check off in logic's bounce.

this can only help the situation.

most likely not why u hear distortion or seein lotta clippin'......whenever u convert to mp3 with such minute headroom like the one ur talkin about, u are gonna have the waveforms exceed the actual peak sample values and will create the inter-sample peak phenomenon....just lower your peak at -1db ...if ur still hearing distortion then maybe it's something to do with dithering but i doubt it.....

[domokunrox]

Originally Posted by aemaury

unfortunately not all of us have years and years of wise, multifaceted expertise on the subject. i come here to learn! certainly i feel trapped in the box... do you find PT to write better WAVs? i cant wait to own a console....

anyway, thanks. i'm gonna try a tad higher of a high pass. SOS/TOMB forums also suggested that i bring the peak down to -.03 or .05.

although, that leads me to a new question... most commercially distributed, professional records peak clean at 0 dont they? i feel like i see that all the time when loading reference tracks in logic. i shouldnt HAVE to sacrafice my mix peak to solve this. its gotta be a frequency content thing.

Its not that PT writes better WAVs. PT stereo sums up better then most DAWs out there. They all sound different in quality, stereo image, and how you choose to make the stereo WAV.

There are differences between PT LE and HD, and there are differences between file > bounce to disk and Internal laybacks.

Don't get too excited to own a console. Consoles are overrated and people care more about the quality of your converters, clock, and front end gear.

As far as peaking clean at 0dB. Nobody should be mixing that hot for a ME. Leave some headroom for a ME to work with. Try aiming to mix around -18dB FS on average.

If I were you, because you're on Logic, I would mix down discrete left and right channel WAVs, have him bring up the (insert loudness word of your choice here), and sum it up for you.

Good luck

[Nordenstam]

Intersample peaks

[taturana]

Originally Posted by domokunrox

Its not that PT writes better WAVs. PT stereo sums up better then most DAWs out there.

except , of course, the ones that use 64 bit FP internal mixing...

[kjg]

Originally Posted by domokunrox

Its not that PT writes better WAVs. PT stereo sums up better then most DAWs out there. They all sound different in quality, stereo image, and how you choose to make the stereo WAV.

it sums better? really? null tested that?

There are differences between PT LE and HD, and there are differences between file > bounce to disk and Internal laybacks.

and null tested that? i'd be interested to hear what you found exactly.

[Alexey Lukin]

It'sJoeAgain and Lupo: the phenomenon of inter-sample clipping is different from clipping happening in mp3.
Inter-sample clipping is clipping of the reconstructed analog waveform between non-clipped digital samples.
Mp3 clipping is a clipping of the digital waveform decoded from mp3 format, it happens because of lossy encoding.

[Nordenstam]

Thanks for keeping check on the quick and flimsy statements! Have been writing about this on the forums for years and gotten a bit tired of it by now..

There's absolutely no doubt that you know this stuff much better than anyone of us will ever do. What I meant was to expand on Joes statement as the issues are related.

Did a quick test with a typical loudness war file with intersample peaks. Did one mp3 pass at 256K with intersample peaks kept as is and another mp3 pass with the IS peaks removed using your limiters "prevent IS peak" option. The file with IS peaks had +1.5dB after the mp3 coding and decoding, while the file without had +1 dB after decoding.

My own experience is that the processing options that produces IS peaks are the most troublesome for the MP3 coders. If I stay well away from such processing, making a master file with real/reconstructed signal peaks at -0.5, the resulting mp3 coding fares very well with little or no clipping. (of course, depending on encoding rate) Whenever I smash the tops with the loudness hammers without any regard to IS peaks, the problem escalates.


Still, you're right, it's not the same thing. My apologies for being too quick..

[Alexey Lukin]

I agree, these 2 problems can add up together.

[Adam Dempsey]

Originally Posted by aemaury

i'm gonna try a tad higher of a high pass.

Originally Posted by Alexey Lukin

Any lossy encoding changes peak signal levels, every peak can randomly become higher or lower. This often leads to clipping, unless some headroom is left.

Additionally, depending on the signal, the ripple from a high pass filter engaged within an mp3 encoder can also be enough to spill signal into the red.

But yes, if I had to choose, I'd go 320kbs AAC over mp3 any day. If I had to data-compress.

[Lagerfeldt]

Originally Posted by domokunrox

Your problem is bouncing masters out of Logic.

You think their WAVs sound good? They make my ears want to commit suicide.

It has this weird, smeared, distorted low end characteristic. Might sound good for Miami booty bass records.

Nonsense.

Originally Posted by domokunrox

Its not that PT writes better WAVs. PT stereo sums up better then most DAWs out there.

Nonsense.

If I were you, because you're on Logic, I would mix down discrete left and right channel WAVs, have him bring up the (insert loudness word of your choice here), and sum it up for you.

Nonsense, and again it wouldn't make any quality difference if you use split files or interleaved. Ever heard of a null test? And pan law?

-
Now, to answer aemaury's question, here's what to do exactly:

Bounce a 24 bit interleaved WAV file (with POW-r#3 or other dither/noise shaping if you like), with a ceiling of -0.3 dBFS, avoiding intersample peaks. There are several (free) plug-ins out there that can show you those.

Use iTunes with the following settings to make an MP3 from the 24 bit WAV file:

320 kbps
No VBR
Joint Stereo
No 10 Hz filter (as Adam says, enabling this can cause overshoots because of phase shifting)

That will give you an MP3 with the sound closest to the original WAV file and with minimal artifacts. Naturally, making a less limited or clipped master can only improve matters. If you don't want to use a ceiling of -0.3 dBFS and instead use a ceiling of 0 dBFS, then you at least need to make sure there are no intersample peaks.

Also, remember to turn off the "sound enhancement" and "sound check" in iTunes when playing back MP3s.

[danika]

Originally Posted by aemaury

i'm mastering with voxengo elephant as the final plugin in the chain. mix reads under 0db without question in logic and my loudness/small dynamic range isnt earth shattering... what is happening? what is the root of this problem! the mix is great and the EQ curve is quite standard and comparable to much of todays music. WAV sounds awesome. MP3s suck on really high quality settings and a "reputable" encoding algorithm

What output ceiling are you using? Did you look at the mp3 ouput in an audio editor and/or do a clipping analysis? There is no limiting in the mp3 conversion, so unless your ceiling is at least -1 to -2 db, you're going to have clipping on transients.

[Thor]

Sonovo-MBP:Desktop Thor$ lame "Great Song Master.wav" --bitwidth 24 --clipdetect --vbr-new -V2 -b 192 "Great Song Master.mp3"

When the job is done, you'll get a number to use with the --scale option to prevent clipping and distortion. Then you plug this number in and re-run the MP3:

Sonovo-MBP:Desktop Thor$ lame "Great Song Master.wav" --bitwidth 24 --scale 0.84 --clipdetect --vbr-new -V2 -b 192 "Great Song Master.mp3"
LAME 3.97 32bits (www.mp3dev.org - mp3devÂ*)
Using polyphase lowpass filter, transition band: 18671 Hz - 19205 Hz
Encoding Great Song Master.wav to Great Song Master.mp3
Encoding as 44.1 kHz VBR(q=2) j-stereo MPEG-1 Layer III (ca. 7.3x) qval=3
Frame | CPU time/estim | REAL time/estim | play/CPU | ETA
7453/7453 (100%)| 0:11/ 0:11| 0:13/ 0:13| 16.712x| 0:00
32 [ 1] *
192 [5947] %%%%%%**************************************************************
224 [ 807] %%%*******
256 [ 499] %%%***
320 [ 199] %**
-------------------------------------------------------------------------------
kbps LR MS % long switch short %
203.1 13.0 87.0 91.1 5.1 3.8
Writing LAME Tag...done
ReplayGain: -6.9dB

The waveform does not clip and is at least 0.1dB away from full scale.

For this particular client a bitrate of 192 was specified (although we were allowed to use VBR instead of CBR for slightly better quality). Obviously you use whatever you can get away with for the best possible result.

Cheers,
Thor

[Jesse Graffam]

Originally Posted by Thor

Sonovo-MBP:Desktop Thor$ lame "Great Song Master.wav" --bitwidth 24 --clipdetect --vbr-new -V2 -b 192 "Great Song Master.mp3"

-b doesn't do anything if you specify a -V. Also there should be a space between -V and 2. --vbr-new gets used by default since 3.97b and later. Also 3.98.2 fixes problems with CBR, and has a few slight improvements for VBR too, you should be using it. It also detects the bit depth of accepted input formats, so --bitwidth is not needed. The output mp3 is also the same name as the input with a changed file extension, so that's not needed either.

lame "Great Song Master.wav" --clipdetect -V 2 -q 0

That would be a better command-line to try using. The -q 0 sets the quality of the encoding to maximum. Personally I use -V 0 though, which averages to a bit under 256kbps for most of what I encode. With 1TB external drives selling on Buy.com for $99 bucks now, there's no reason not to max quality. (or just use FLAC, for that matter )

Originally Posted by Thor

For this particular client a bitrate of 192 was specified (although we were allowed to use VBR instead of CBR for slightly better quality)

Actually it is possible to do a bit-rate defined version of VBR, called ABR... where it tries to use the bits where it can inteligently, but it keeps the relatively short-term average bitrate whatever you intended. ABR is not recommended because the quality isn't as good as VBR, and many times you end up near the same size with a much better quality in VBR. ABR can force the encoder to use more bits where it sometimes isn't needed, so that's part of the problem when compared to straight quality VBR using the -V string. (which ignores -b)

The target average of -V 2 is ~192kbps though.

LAME - Hydrogenaudio Knowledgebase

[PatrikT]

Originally Posted by aemaury

WAV sounds awesome. MP3s suck on really high quality settings and a "reputable" encoding algorithm

If the wav is properly balanced form a mastering point of view, for it's own good off course, then the mp3 won't suck. There's just no chance. There will happen things, along with a shrunk bandwidth, but they should not cause any major state of shock.

As little as Logic suck, because it doesn't either.

CPH rules, by the way.

Audio is highly operational-resulted and it has always been popular to blame all kinds of things except oneself.


Best Regards
Patrik

[Thor]

Some good points here.

I haven't been able to find a binary of 3.98.2 anywhere and don't have the developer tools installed to compile it myself. So 3.97 will have to do for now. There have been some developments, as you note, but we are talking MP3 here....

While I generally agree that quality should be max, it's also up the the client, and many of them have specific target file sizes for downloads, etc. It's not a question of disc space per se, as far as I know. So we find a happy medium.

Cheers,
Thor

Originally Posted by uncajesse

lame "Great Song Master.wav" --clipdetect -V 2 -q 0

That would be a better command-line to try using. The -q 0 sets the quality of the encoding to maximum. Personally I use -V 0 though, which averages to a bit under 256kbps for most of what I encode. With 1TB external drives selling on Buy.com for $99 bucks now, there's no reason not to max quality. (or just use FLAC, for that matter )


The target average of -V 2 is ~192kbps though.

LAME - Hydrogenaudio Knowledgebase

[doug hazelrigg]

Originally Posted by domokunrox

Your problem is bouncing masters out of Logic.

You think their WAVs sound good? They make my ears want to commit suicide.

It has this weird, smeared, distorted low end characteristic. Might sound good for Miami booty bass records.

I agree, this is nonsense. A wav is a wav

Its not that PT writes better WAVs. PT stereo sums up better then most DAWs out there. They all sound different in quality, stereo image, and how you choose to make the stereo WAV.

Again, just nonsense

[smoke]

Originally Posted by Lagerfeldt

Nonsense.


Nonsense.


Nonsense, and again it wouldn't make any quality difference if you use split files or interleaved. Ever heard of a null test? And pan law?

-
Now, to answer aemaury's question, here's what to do exactly:

Bounce a 24 bit interleaved WAV file (with POW-r#3 or other dither/noise shaping if you like), with a ceiling of -0.3 dBFS, avoiding intersample peaks. There are several (free) plug-ins out there that can show you those.

Use iTunes with the following settings to make an MP3 from the 24 bit WAV file:

320 kbps
No VBR
Joint Stereo
No 10 Hz filter (as Adam says, enabling this can cause overshoots because of phase shifting)

That will give you an MP3 with the sound closest to the original WAV file and with minimal artifacts. Naturally, making a less limited or clipped master can only improve matters. If you don't want to use a ceiling of -0.3 dBFS and instead use a ceiling of 0 dBFS, then you at least need to make sure there are no intersample peaks.

Also, remember to turn off the "sound enhancement" and "sound check" in iTunes when playing back MP3s.

Lagerfeldt, I enjoy your tips. What do you consider are the differences between joint stereo and normal stereo in iTunes. I mean in layman's terms your opinion on the sonic or spatial difference in the resulting file?

[Jesse Graffam]

Originally Posted by smoke

Lagerfeldt, I enjoy your tips. What do you consider are the differences between joint stereo and normal stereo in iTunes. I mean in layman's terms your opinion on the sonic or spatial difference in the resulting file?

The way I look at it, iTunes should not be used for encoding, no matter what. It uses FhG (Fraunhofer Group) HQ codec for encoding, instead of F.A.S.T. which is FhG's current reference codec. They have also turned down the quality of iTunes' HQ codec so that it competes with the encoding speed available with Real Player and others, which are mostly using Helix (Real), LAME, or F.A.S.T.. FhG doesn't give their layer-3 coding much thought these days because it is working on a TON of other things, and so they have fallen behind, especially compared to LAME which actually tests the quality of it's VBR output and makes sure each block is actually high enough quality after decoding to represent the quality factor desired, not just predicting it's output like everything else.

Originally Posted by doug hazelrigg

I agree, this is nonsense. A wav is a wav. Again, just nonsense.

Yes, absolutely @ a WAV being a WAV. But... (and it's a really big butt) a mixing engine is not just a mixing engine. And it's entirely possibly that someone might like the sound of doing what the OP is about, with a specific mixing engine. And without you actually trying this with the exact program & the exact version that the OP's friend is talking about - any of your claims that it is nonsense are bogus.

I'm not saying it's true or untrue, because I have not tested it for myself. The possability exists for this difference of mixing methods to make a difference resulting in "better" sound quality by way of transparency, weather you agree or not.

Would the mixing engine have to be screwed up in some way for this to actually work? I'm not totally sure about that either. It might come down to the dithering method, and the possibility that it really does sound more transparent being dithered in multiple stages like this, with the input tracks to the second stages having an LSB below the final WAV's LSB within the mixing engine. It might not.

Originally Posted by Thor

I haven't been able to find a binary of 3.98.2 anywhere.

RareWares
click on: MP3 -> LAME Bundles, and 3.98.2 is second on the list.
Just in case anyone is tempted, NEVER use any Alpha builds of LAME. The main reason is they are trying to improve things, and those are available for the many testers (and experienced at listening for coding artifacts) to be able to confirm if the difference was good or if it was bad, sometimes with very specialized audio samples to test certain things. So don't use any versions of LAME that are not official, and confirmed to be awesome.

[Lagerfeldt]

Originally Posted by smoke

Lagerfeldt, I enjoy your tips. What do you consider are the differences between joint stereo and normal stereo in iTunes. I mean in layman's terms your opinion on the sonic or spatial difference in the resulting file?

You're welcome!

I'd love to explain it myself but someone else already did this for me:

MP3 conversion: Joint Stereo - The Myths And The Realities
Joint Stereo: The Myths ... and The Realities

Rather long, but very interesting. It's a question of joint stereo improving over-all sound quality, not just spatial information, by using the M/S principle. Correctly implemented joint stereo (not to be confused with intensity stereo, which is sometimes referred to as joint stereo too) will always give you the best sound quality since it switches between M/S and regular stereo encoding whenever necessary. This way you get the best of both worlds.

[Thor]

Windows only.

I do see a 3.98 for Mac OS X further down though, and will try it out. No 3.98.2 for OS X, and as I said I don't have the dev tools installed.

Thanks for the URL.

Thor

Originally Posted by uncajesse

RareWares
click on: MP3 -> LAME Bundles, and 3.98.2 is second on the list.
Just in case anyone is tempted, NEVER use any Alpha builds of LAME. The main reason is they are trying to improve things, and those are available for the many testers (and experienced at listening for coding artifacts) to be able to confirm if the difference was good or if it was bad, sometimes with very specialized audio samples to test certain things. So don't use any versions of LAME that are not official, and confirmed to be awesome.

[domokunrox]

Hold the presses here.

My last reply in this thread has been replied at with a nonsense nonsense sort of fashion.

Are you people seriously calling it that on summing sounding different for other DAWs? Consoles too!

Can you hear the differences between summing on NeveVR ? SSLG? SSLJ? PTLE? PTHD? PTMP? Nuendo? Logic? APIxxxx? Sonar? DM2000? Live?

They all don't sound the same, guys. You're kidding yourself if you don't think there is a difference between a PTLE internal layback and PTHD internal layback.

[Jesse Graffam]

I don't think most people disagree with you that there are differences in some cases, as far as how a mix is "bounced" to 2-track. And in some cases they may also be audible to some people.

But it's only helpful to people if they try it themselves and are able to ABX it, or if someone else does, proves they can ABX it, and provides information as to the exact version numbers of the software in question.

The topic was never about the difference between mixing engines or devices.

[edit]
nevermind, i'm replying to the wrong topic... the bouncing & re-mixing sub groups vs bouncing all at once topic. MY BAD.
[/edit]