MP3 vs WAV: formal study?

[Manning]

I am quietly delighted to have provoked so much debate. It's raining like hell in Sydney so we all need something to do.

[Hookemeister]

If you really want to hear what MP3 encoding is doing, try this...

Only prerequisite is having a way of putting one channel out of phase.

1: Grab a CD and encode it at various bit rates.

2: Take the output of your MP3 and feed it into 2 channels of your console or DAW.

3: Centre the panpots on both channels.

4: Open the fader on the left channel.

5: Hit the Phase switch on the right channel or put the right channel out of phase.

6: Slowly fade up the right channel until all the centre image (Vocal, Kick, Bass) disappears.

What you're doing is nulling anything that is of equal level in both the left and right sides of the mix, in other words anything that's mono and panned centre.

What you'll hear is only the stereo component of the mix in mono, and the artifacts of the MP3 encoding. It is particularly noticeable on hi-hats and cymbals especially on low bit rates,

7: Now try the same thing with a CD and I think you'll notice the difference.

The above method has another valuable use: It is also an excellent way of listening to what's happening in the vocal reverb department of your favourite CD as the vocal is usually centred and the reverb is usually in stereo. Once the vocal is removed you get to hear the delays and reverbs on their own.

I went thru a period in the 70's, where everytime I bought a new album, I would listen to it once normally then phase invert one channel and listen to it with all the mono image removed. I learnt a lot about mixing from that.

Try it... it's fun.

[Text_Edifice]

Or, take your mp3 and pcm file, phase invert one and combine them together. That will give you the 'difference' between your CD and mp3 version.

[ChrisW]

The sad reality is that an mp3 is extremely convenient. Even many musicians I know are using mp3's on a daily basis. I've met up with guys who play me stuff they've recorded on their iPhone mic. Probably because it's the music that matters, and capturing the moment with whatever means we have at hand, and is convenient.
Now the above doesn't mean we should aspire only to the average standard in our work.
It isn't an either/or. Everyone I know has embraced both - higher bit rate recording, consumer quality playback.
There is little difference between WAV and mp3 when I'm listening to music in my car, or on earbuds in a noisy environment. When I'm in a dedicated listening space, I want to hear the best quality.

[Drumstruck]

I am quietly delighted to have provoked so much debate. It's raining like hell in Sydney so we all need something to do.

You should be very proud of yourself Mr Manning - it's rare to see the t'rockers so vocal (even if some of us are a little off topic).

* Green zones - great idea
* Size of Oz (compared to piddly little o/s countries) contributing to poor comms - very true but I wouldn't change that for any money
* Would I give up living (a bit) remotely for faster comms? snowballs in hell come to mind..... despite my bandwidth dropping to dialup speed yesterday due to me using all the download capacity (what's that word that rhymes with truck??)
* mp3s convenient for cars etc - indeed true - so much ambient noise in cars that dynamics are pretty useless (I don't own a bentley....)

and back on topic (somewhat)

So the future gents? (after the formal study) - what is the real prediction?

The major hurdles (impacting moving from mp3 to wav) appear to be fast comms available everywhere and the unknown user future demand...... will the listening public want a file format with greater dynamics or are we all doomed to compression?

And will this push human evolution into a being with lesser audio perception?

[Kurt]

You appear to be mixing your terminology up. File compression has absolutely nothing to do with dynamics.

[smash]

A couple a slight technical things...

1. iTunes doesn't sell MP3's. Never has. Originally they sold 128 Kbit AAC files, then they switched to 256 Kbit AAC files, iTunes plus. Recently, they have introduced "Mastered for iTunes" which although being only 256 Kbit AAC is encoded in 32bit floating precision directly from whatever format you give it.

Call me a "fanboi" but Apple has done more than many other companies to push the quality of digital audio - including their answer to the loudness war - Sound Check.

It's quiet easy to say "just sell everything uncompressed" but it isn't iTunes you need to complain to - blame the labels!
It is still the record labels who dictate the maximum quality that the various online stores let you sell things at.

http://images.apple.com/itunes/mastered ... itunes.pdf

2. YouTube are actually fantastic with the format you can upload. They allow files greater than 2GB, and will take 5.1 audio, uncompressed, as well as uncompressed video or other professional mastering formats like ProRes 444 and AVID DNX HD.

http://www.reduser.net/forum/archive/in ... -70894.htm

The key with all digital media, you will get the best results by using the lowest possible number of compression passes prior to your final output.

Disclaimer: Apple and YouTube both give me money to do stuff occasionally.

Simon "Smash" Ashby

[Alastair Reynolds]

As an aside, has anybody done any work with OPUS codec yet ?
From my point of view, low delay and fullband for remote broadcast looks interesting.This would also depend on how FEC (forward error correction) could be implimented whilst maintaining minimal delay. Both mp3 and AAC have long coding delays which make real time audio difficult for broadcast use.

[Drumstruck]

You appear to be mixing your terminology up. File compression has absolutely nothing to do with dynamics.

No, it's late in the day, stream of consciousness typing and assuming some reading between the lines as I wax philosophical.

Consider my balls busted and I shall say no more - dummy spat

[rick]

A wav is something that hits you in the chest at the beach
Not sure what an mp3 is but there are spaceships involved
For sure


The chorus is the part of the song you sing
Along with untill you learn the verse and all the
Hooks ..

If I am wondering about the quality of a mix I get
( mp3.. Are curiously hard to eq )

I mono the signal and flipp the phase On one side
All the aliens get out of their spaceships
And shout out "I surrender .. Yes you have identified us
We are not from here "

Ymmv ..if you can't hear as well as me or don't know much about mono
Spaceships
or
phase

[Paul Maybury]

Hi Alistair, how are you dealing with any MetaData requirements now that you are storing audio as PCM's?
here at PBSFM we are trying to institute digital playout systems and keep coming up against the fact the our preferred playout system (MegaSeg) requires MetaData to function. So we are looking at Apple Lossless.
The death of the CD is all too obvious from a radio point of view. Labels are no longer providing CD copies of new releases. Even if they are, it will be only one or two copies. (we have over 80 announcers). These will often just be burns, and who knows what encoding has gone on with these.
For stations with limited playlists and little concern for audio quality other than being the loudest on the dial, (I'm looking at you Austereo. Oh, and JJJ...so loud and dense at only 80% modulation..) storage of library audio is fine as MP3's I suppose,
but for us, we have an enormous library and prefer to archive at the highest available quality. At the moment, that is still 16 bit 44.1khz PCM for commercial releases.
MP3's do not take well to further processing, whether it be the broadcast processing, or in the production suite.
So it looks like one of the "Lossless" encoders is the best option at the moment.
Somewhere in my reading on this debate was someone claiming to hear a difference between AAIF and WAV files of the same audio....
Cheers,
Paul Maybury

[smash]

Post of the year from gearslutz about the difference between WAV and AIFF:

Originally Posted by Vortex4001
YES, YES, AND YES!!!! AND ANOTHER YES. There is a sonic difference between AIFF and WAV, Wav has more low end and dynamics, AIFF was invented for broadcasting systems by apple, so it adds a little compression to the low end, and smooths out the Mids and High end, Therefore it is a little more pleasing to the ears when used for a particular genre, such as pop which is basically smoother sounding than any other genre anyways due to increased compression. So I suggest if you're gonna mix Hip/Rock>Go with wave>For Jazz, Country and RnB, go with AIFF. trust me there is a difference. Thats why alot of industry engineers use both formats depending on the genre they're mixing.

And word of advice, Just because you post a question on a forum, dont expect a 100% accurate answer, alot of people are puffed up with arrogance and Self-conceitedness, that they feel they know-it-all and will just blurt out anything they feel is right. ALWAYS GO AND RESEARCH THE TOPIC FIRST BEFORE POSTING IT ON A FORUM.

In summary: AIFF is richer, WAV is more aggressive.

That guy should win a comedy award!

Simon "Smash" Ashby
If you build it, we still won't come.

[mylesgm]

classic gearslutz right there... "research the topic before posting on a forum"

[DarkSky]

That guy should win a comedy award!

The fact that this bucket of comedic tripe was posted on 1st April 2012 is probably the hidden "clue" to what would otherwise seem an outlandishly clueless post.

[Alastair Reynolds]

[quote="Paul Maybury"]Hi Alistair, how are you dealing with any MetaData requirements now that you are storing audio as PCM's?
here at PBSFM we are trying to institute digital playout systems and keep coming up against the fact the our preferred playout system (MegaSeg) requires MetaData to function. So we are looking at Apple Lossless.
The death of the CD is all too obvious from a radio point of view. Labels are no longer providing CD copies of new releases. Even if they are, it will be only one or two copies. (we have over 80 announcers). These will often just be burns, and who knows what encoding has gone on with these.
For stations with limited playlists and little concern for audio quality other than being the loudest on the dial, (I'm looking at you Austereo. Oh, and JJJ...so loud and dense at only 80% modulation..) storage of library audio is fine as MP3's I suppose,
but for us, we have an enormous library and prefer to archive at the highest available quality. At the moment, that is still 16 bit 44.1khz PCM for commercial releases.
MP3's do not take well to further processing, whether it be the broadcast processing, or in the production suite.
So it looks like one of the "Lossless" encoders is the best option at the moment.

Addressing these in order
Metadata in PCM files can either be in the Cart Chunk of a BWF (Broadcast Wave File) or in the case of Fairfax radio, Austereo, DMG etc DAF files from DCS Maestro (Austereo may have updated since I last saw) ARN uses NextGen as do many regionals, ABC uses Dalet (I think) Michael Blackburn may shed some light here, some other regionals use Maximation and there are plenty of others to choose from in the Pro market, e.g. Dira-Radio from Germany ,Wide Orbit Automation for Radio, RCS Master Control.
DAF Files are basically Raw PCM but reversed from WAV (Little Endian / Big Endian --- Google it) with a 256 byte header containing the timing, artist ,track , key numbers, Agency ID, secondary, tertiary and EOM info, expiry and deletion dates and a few other bits and pieces.
Playback is through professional Audio cards, Ususally Audio Science or Digigram or Orban but there are a few others around.These cards usually contain a few DSP Chips and high end DAC etc to allow hardware decoding of encoded audio where neccessary without placing any load on the host processor/ GUI.
We used to encode MP2 Audio at 256K Joint Stereo when Hard disks and Raid arrays were expensive, e.g. $20000 for 70GB Raid 5 back in 2000, before that when a 1GB drive was the ants pants we used 8 bit uLaw, to give us about 70dB S/N. Not spectacular by today's standards but better than tape and lets face it ...on AM with about 50dB S/N it was fine.
Now a Professional Playout workstation like this is going to set you back at least $10,000 per node. About $4k for a basic Audio card , $3-4K for a client licence and at least $3K for Computer hardware with enterprise grade mirrored disks. These things are on 24/7 and must be extremely reliable. So Stripped down / locked O/S and nothing else at all, No Antivirus, Internet ....zip .
Central servers to manage distribution between nodes so all nodes have the same set of carts and lock Carts that are queued for play, so another workstation can't change a cart accidentally that is about to go to air.

"For stations with limited playlists and little concern for audio quality other than being the loudest on the dial, (I'm looking at you Austereo. Oh, and JJJ...so loud and dense at only 80% modulation..) storage of library audio is fine as MP3's I suppose"

That's a pretty strong opinion given the investment in systems and equipment by these organisations. Austereo would have spent at least $250,000 on their playout system in each station so for a community station looking at spending $199 US on playout system software to run on a Mac I wouldn't be knocking them.

Megaseg appears to be a program aimed at the DJ setups or Community/ Part 15 (FCC) amateur stations .
Apple Lossless is not a recognised broadcast standard, nor is 44.1K Sample Rate. It is a consumer format.

On board audio or consumer grade cards are not used, the performance is just not good enough.


Austereo does not store any music on their playout system as mp3. Years ago(Like Mid 90s) they ran APTX at 32K sample rate , but like us, transitioned to linear as storage became cheaper.
Nova (DMG Sations) have been Linear PCM , using Maestro DAF format since they started. The original storage arrays were huge, with a huge price tag.

Loud and Dense is what suits their format and listeners. Modulation will be tightly controlled to avoid distortion and mixing products when hey are combined with other stations at the transmission site. They will also be using multiple ACS (Auxilliary carriers) around 67KHz and 92KHz for RDS and OB Cue lines, so again tight control is required. Plus ACMA watches closely and no-one wants a call from them because you're overmodulating.

There's a bit for everyone to think about

[Paul Maybury]

Thanks Alistair for a very comprehensive reply. Lot's of helpful info there.
And food for thought.
You are right of course about processing levels being appropriate for the target markets. For the commercials it makes sense to have a "station sound". Anyway, that's a whole other argument.
Obviously I am behind the times with my information about how the commercials store their audio. I find it encouraging that as hard drive space gets cheaper, stations are using it to store higher quality audio. Now, if consumers would follow that lead.
Digital library playout is new for a station like PBS. The desire to get it operational is a direct result of CD distribution dying away.
We have an entirely different system for cart playout and scheduling (DineSat. It is horrible!). Mega Seg was chosen for music library playout because it is familiar to announcers, being a D.J. style GUI. We don't use onboard soundcards.
Community radio is about 10 years behind the rest of the broadcasting world.....I wish that wasn't the case.
Cheers,
Paul Maybury

[jasound]

Awesome post Alister.
This is why I'm a turtlerock forum member.
I might have to read it again though.

[Drumstruck]

Awesome post Alister.
This is why I'm a turtlerock forum member.
I might have to read it again though.

+1 excellent reading

[Paul Maybury]

Hi anyone still reading. I thought I'd clarify the muddy waters I seem to have stirred up.
Firstly, everyone I've met who works at the tech or production end of radio wants it to sound good and feel they have worked hard to achieve that.
So I don't want to dis someones work, and I certainly didn't intend to start any Us vs Them silliness. We all care about audio, that's why we are here.
Fringe area coverage, station sound (marketing) and density are all drivers of heavy on air processing. This is the same for every station. I understand why it is done, but on most commercials I really find it fatiguing. JJJ even though public are probably the worst of all, and they don't use all of their modulation allowance. Now, that is odd.
The $250k Alastair mentions being sunk into every station for digital systems by Austereo is of course a significant investment. These systems often combine scheduling and reporting, plus automated playout. These features are an asset to any station selling airtime.
By contrast $250k would cover a complete refit of PBS from turntables to STL. Maybe even the transmitter too.
We still run an entirely analogue broadcast chain. We have turntables, cd players, cassette players and even minidisc still in our on air studios. Even our STL is still analogue. So yes, we are running a station with amateur budgets and some consumer equipment, such as turntables and cd players, into analogue consoles.
The point is, even at this level, MP3 audio suffers badly by caparison to even 16 bit 44.1khz cd audio. By the time the audio reaches your receiver, it has traveled a long way through equipment that would make an audiofile feel queezy and mildly shock most studio engineers. The best hope for the audio is to start with the best resolution available, do not encode/ decode it any more than you have to and try to convince management that we don't need to choke the life out of it with the broadcast processing.
Maybe it would be good to start another thread about broadcast issues for audio engineers,
anybody keen to chime in?

Cheers,
Paul Maybury

[justmull]

Great insight into the broadcast world. Yes please don't do any re-encoding radio stations - I suspect that is why my local community broadcaster has recently become unlistenable.
Subjectively I can say that with MP3 at high bitrates the biggest impact on the sound is whether or not you encode in Joint Stereo which is the default for most codec implementations. You have to dig into the encoding options to find this if the codec even gives you the option. Joint Stereo takes advantage of any similarities between the L and R channels to reduce filesize. It's what I think is the biggest contributor to the "watery" effect people describe. I notice it most on the hi-hats. Encoding at 320kbps True Stereo I will admit that I have a hard time distinguishing "common content" (record and cd rips) wav files from mp3 files when listening on consumer equipment. Sound Forge has the option to encode in True Stereo though there may be other implementations (I know of at least one).

[Paul Maybury]

So that's what "Joint Stereo" means! Thanks Justin.

[Alastair Reynolds]

Joint stereo is a way of delivering a wider audio bandwidth with a narrower degree of spatial information.
It works well with typical music sources. We use it with mp2 based hardware encoders to give us 20KHz audio bandwidth in a 128K data channel with some spatial information on what would otherwise be a mono feed.

if we used discrete Stereo or Dual Mono , the bandwidth would be halved (10KHz) for each audio channel, but of course the separation is vastly improved. We use this Dual Mono mode for Outside broadcast where separation of program/Cue/Order wire lines etc are needed, but 10K is adequate bandwidth for voice only contribution (especially for an AM station in glorious Lo-Fi Stereo-monic).
128K is usually about it for OBs , particularly on ISDN lines as bonding/aggregating multiple ISDN lines to give a single data capacity of 256K or 384K gets tricky with timing and routing through the public telephone network. It can be done , but it's a lot easier to run on a single pair of bonded lines at 128Kbps

The AAC and AAC+ algorithms however use parametric stereo, which is a further developement along the joint Stereo idea.
Much smaller amounts of Data are needed to give a spatial effect which is good enough for the average consumer.
http://en.wikipedia.org/wiki/Parametric_Stereo
Start here for an idea, after that "Google is your friend"

It's a bit like an M/S encoding with only 2-3Kbps of side information to work on a mono signal. This combined with its partner in crime SBR or Spectral Band Replication which basically synthesizes everything above 10KHz in order to reduce bandwidth and you get the picture of how crappy things can soundhttp://en.wikipedia.org/wiki/Spectral_band_replication it's also used in MP3Pro and some MP3 implimentations.

DAB+ (digital Radio) uses AAC+ to allow Commercial stations to cram more stations onto our legislated allocation of 128K per existing licence.
Plenty of Music stations are transmitting at 48Kbps Parametric Stereo, with SBR and using between 4-8K of this as Slides Show and text (title artist news updates tweets blah blah blah)so the Audio payload may be only 40kbps.
Your Typical Community station allocation is 64K.
Which makes for interesting discussion as under the legislation , the Commercial stations had to build and pay for the complete DAB infrastructure and GIVE over 20% of our Bandwidth to community stations, although we are allow to charge a market rate to use it. Still, on a $10 Mllion per city build it will take us a long time to recover our costs, but that's another story.

For Sound quality on 2UE's DAB transmission I elected to use discrete Stereo and no SBR and restrict the audio bandwidth a little. (about 13KHz)I run 80kbps with about 4kbps for Slideshow/Text, so around 75kbps audio payload. After extensive listening we found this to be less fatiguing to listen to on voice programs . Highly processed music stations can mask the artifacts a lot more easily than cold voice so they get away with lower bit rates. Austereo, to their credit, bought additional bandwidth at considerable cost and transmit their main stations at around 96kbps.
I did run 2UE for nearly 2 years at 128Kbps and it sounded very good for DAB, but had to add anothe service and sacrificed 48K of my available bandwidth.
FM Transmission by the way is band limited to 50Hz-15KHz, AM is 50Hz to 7.5KHz in Aus, 8.5KHz in US (NRSC spec)

It's all fun isn't it...

So my vote always goes for PCM with no data reduction . To my ears it sounds better (for digital)

[Paul Maybury]

Thanks Alastair, great reading!

[Sammas]

Not a formal study... but interesting none the less.

http://www.trustmeimascientist.com/2013 ... t-everest/

[Kurt]

We can't hear the difference but I can hear the crickets...
It is very interesting, I wonder how long it will take for someone to successfully take the challenge.

Not a formal study... but interesting none the less.

http://www.trustmeimascientist.com/2013 ... t-everest/

[Mark Bassett]

So has anyone ever seen a formal, double blind test which settles the matter? There's anecdotal evidence everywhere, but that doesn't settle arguments.

As far as tests with reported data go, Dr Sean Olive from Harman published on this in 2011 in the AES. 'The LAME 3.7 (version 3) MP3 codec was used in these tests set at a fixed bit-rate of 128 kbit/s'.

Olive. S.E. (2001) 'Some New Evidence That Teenagers May Prefer Accurate Sound Reproduction' Presented at the 131st AES Convention 2011 October 20–23 New York, NY, USA.

Abstract:
A group of 18 high school students with no prior listening experience participated in two separate controlled listening tests that measured their preferences between music reproduced in 1) MP3 and lossless CD-quality file formats, and 2) music reproduced through four different consumer loudspeakers. Overall, the teenagers preferred the sound quality of the CD-quality music reproduction, and the most accurate, neutral loudspeaker. Together, these tests provide some new evidence that teenagers can discern and appreciate a better quality of reproduced sound when given the opportunity to directly compare it against lower quality options.


Conclusion:
This study was partially motivated by the current popular myth – reported widely in the media - that young consumers are either indifferent to the quality of reproduced sound, or they actually prefer bad sound to good sound [2], [3]. The truth is that there exists very little scientific evidence to either support or refute these claims. To further shed some light on this question, two controlled, double-blind listening experiments were conducted to determine the sound quality preferences of 18 high school teenagers with no prior experience in critical listening.

In the first experiment, the teenagers made preference choices between four program selections presented in both MP3 (128 kbit/s) and lossless CD-quality formats. The teenagers preferred the higher quality CD-quality music reproductions to the MP3 versions in 67.1% of the trials, a statistically significant result (p < 0.05). There was no statistical evidence that any of the listeners preferred the lower quality MP3 versions.

http://www.aes.org/e-lib/browse.cfm?elib=16108

[Text_Edifice]

Thanks Mark,

that's interesting. I'm unsure whether a study with 18 participants can generate a 'proper' statistically significant result (as opposed to a p-value <.05).

I'm sure this is addressed in the paper proper, will download it when I get to work.

[Manning]

128 KBit is hardly a standard MP3 bit-rate anymore is it?

[Manning]

I have also observed a statistically significant correlation between rain in Sydney and the resurrection of this thread

[Sammas]

128 KBit is hardly a standard MP3 bit-rate anymore is it?

I don't think you can buy it anywhere. I know Itunes doesn't offer it, and the minimum beatport offers is 320kbps.
Judging mp3's on 128kbps is kind of like judging PCM audio using 8bit 44.1khz files.

Does appear anyone has actually taken up the offer yet of proving they can differentiate between 320kbps MP3's and .wav files in the link I posted above.