WAV <--> MP3 ... the 'Xerox effect' examined!

Post your questions & get help from friendly LibriVoxers
Post Reply
Cori
LibriVox Admin Team
Posts: 11970
Joined: November 22nd, 2005, 10:22 am
Location: Great Britain
Contact:

Post by Cori » June 12th, 2007, 3:20 pm

So ... I've heard often here that changing files from MP3 to WAV for editing, and then changing back is okay once or twice, but after a while you get the Xerox effect -- that is, the audio quality being eroded like a photocopy of a photocopy of a photocopy ...

And I thought it'd be interesting to hear exactly what that effect sounds like, and how many changes are needed for it to kick in.

http://upload.librivox.org/share/uploads/cs/busybee-final.mp3 (670Kb) has an answer, for your ears too.

I recorded the first stanza of a semi-random poem in MP3 - 192Kb rate, as I usually do, and then converted to WAV and noisecleaned, before tuning down to LibriVox standard MP3 - 128Kb rate. That's the base file.

Then I simply converted the file to wav, and back to mp3 until I got very bored of doing it, which was after 20 repetitions. In the recording, you can hear, in sequence:

The basic conversion.
... that file converted 5 times in total.
... that file converted 10 times in total.
... that file converted 15 times in total.
... that file converted 20 times in total.

Conclusion: the Xerox effect kicks in pretty quickly somewhere between 7 and 10 repeats, and then gets cummulatively worse. If you are about to convert and edit a file of unknown history, listen closely to the original ... if it's showing any signs of degradation, the effect magnifies the more often it's done, and consider alternatives if at all possible. (i.e. editing in the MP3 format, as I've done to glue the files together here, or contacting the reader for preceeding versions.)

However, for those who do this conversion back and forth as a regular part of their editing process, you have more leeway than you might think for changes ... certainly up to five times without a terribly noticable effect, and even as many as ten, if that were bizarrely necessary.

[edit = file link updated]
Last edited by Cori on April 19th, 2009, 7:09 am, edited 1 time in total.
There's honestly no such thing as a stupid question -- but I'm afraid I can't rule out giving a stupid answer : : To Posterity and Beyond!

Starlite
Posts: 16682
Joined: April 30th, 2006, 2:17 pm
Location: Thunder Bay Ontario, Canada

Post by Starlite » June 12th, 2007, 4:10 pm

Wow thats amazing Cori. You must have been really bored. :P Oh ya I can hear the degredations. This should really go into a podcast!! :D

Esther
"Reasonable people adapt themselves to the world. Unreasonable
people attempt to adapt the world to themselves. All progress,
therefore, depends on unreasonable people." George Bernard Shaw

a.r.dobbs
Posts: 3237
Joined: February 23rd, 2006, 1:04 am
Location: Boston

Post by a.r.dobbs » June 12th, 2007, 6:51 pm

Wonderful, Cori!
Your original sound was marvelous, though. I wanted to comment that the back and forth effect can become noticeable earlier on recordings that start out sounding a bit more digital -- as do certain mics on a PC, for instance. I hope to back that statement up with some proof samples in the future.
Anita

Cori
LibriVox Admin Team
Posts: 11970
Joined: November 22nd, 2005, 10:22 am
Location: Great Britain
Contact:

Post by Cori » June 13th, 2007, 12:01 am

If anyone wants to PM me a sample, I'll happily Xerox them for the record!

Not that I want to be deluged, mind, but ... if people are curious about their own setup, or have a more digital sound to begin with, as Anita describes, then it WOULD be interesting to hear a range.
There's honestly no such thing as a stupid question -- but I'm afraid I can't rule out giving a stupid answer : : To Posterity and Beyond!

Gesine
Posts: 14185
Joined: December 13th, 2005, 4:16 am

Post by Gesine » June 13th, 2007, 2:51 pm

Interesting test, Cori. Thanks!
[size=92] "Imagination is more important than knowledge. Knowledge is limited. Imagination circles the world." Albert Einstein[/size]

Stephan
Posts: 1607
Joined: December 18th, 2005, 9:38 am
Location: Leverkusen, Germany

Post by Stephan » June 13th, 2007, 7:16 pm

Thank you so much for this! I always wanted to hear a "Mugshot" of the artifact. Thanks for showing the much larger leeway we have - this is so soothing. It was always like "BEWARE THE KILLER MP3 CONVERSION MONSTER ARTIFACT". HA! Fbhracht!
[url=http://librivox.org/wiki/moin.cgi/PromotionalMaterial][color=indigo]Want to promote LV? Print the poster and pin it at your library[/color][/url] | [url=http://librivox.org/wiki/moin.cgi/Stephan_Moebius][color=indigo]My wiki page[/color][/url]

chocoholic
LibriVox Admin Team
Posts: 13607
Joined: January 16th, 2007, 9:23 am
Contact:

Post by chocoholic » August 21st, 2008, 11:25 am

Resurrecting this thread because the subject has come up several times. Cori has kindly done serial edits/resaves on a couple of my files. My recording setup isn't as good as hers to start with, so this should be a good test.

1. Recorded on Logitech USB desktop mic with no post-processing. Cori recorded a clip of it after 5, 10, 15, and 20 conversions. She increased the volume of each conversion to match the volume of the original (there was a marked volume drop).
http://upload.librivox.org/share/uploads/cs/verne-xeroxd.mp3 (972kb)

2. Recorded on Zoom H2 and noisecleaned (by me) because of some loud background buzzing. Cori recorded clips after 5, 10, 15, and 20 conversions; she didn't adjust volumes on this one.
http://upload.librivox.org/share/uploads/cs/beautybeast-xeroxd.mp3 (1.06MB)

To my ears, up to 5 conversions of an mp3 should be just fine, and more are probably OK too though they might need amping. --Which probably means that except for special cases like difficult background noise, mp3's will be fine even when the files will be edited and resaved (i.e. dramatic works). Other opinions?
Laurie Anne

Cori
LibriVox Admin Team
Posts: 11970
Joined: November 22nd, 2005, 10:22 am
Location: Great Britain
Contact:

Post by Cori » August 21st, 2008, 12:15 pm

The volume-level drop was similar on both files, but when I noticed it on Verne, I thought it'd be more interesting to amplify the results, to hear if there was quality-degradation as well as just a volume loss. I didn't amplify each time I did the conversion ... just once at the end, on the snippets I pasted together.

I don't remember the volume changing much when I did this on my own file. And again, I agree with LA that the effect isn't a big deal up to about 5 formats. If anyone else wants me to test with their mic, please drop a sample file here or PM me. So far we've covered a Logitech, Zoom and Samson. I don't think there's anything innate about a mic which would cause the artifacting to be immediate, but, I'm happy to test!
There's honestly no such thing as a stupid question -- but I'm afraid I can't rule out giving a stupid answer : : To Posterity and Beyond!

chocoholic
LibriVox Admin Team
Posts: 13607
Joined: January 16th, 2007, 9:23 am
Contact:

Post by chocoholic » August 21st, 2008, 4:51 pm

From another thread, in regards to this discussion:
padraigo wrote:Hi.

In terms of what is easy the difference in Audacity for example between creating a .Flac file instead of an MP3 is simply in selecting "Flac Files" in the drop down instead of "MP3 Files" when exporting, which doesn't sound very onerous to me.

The other audio editing tools I've used are similar, Flac and Mp3 are generally just 2 of the available options and it is just a question of choosing the right one when you save.

The amount of loss you get when saving as MP3 depends on many factors so you can't tell in advance if one recording is going to be more affected than another.

If you want .Flac files to open in iTunes you can easily configure your browser to do this one time and then any time you click on a link to a .Flac file it will open in iTunes, just like .mp3 files do (similarly for .wav). If you let me now which browser you use I can explain how to do this if anyone would like to try it.

Additionally the tests referenced simply convert the same file back and forward from .mp3 to .wav. A more realistic test involves actually editing the content between saves which could mean that you suffer different amounts of loss between edited and unedited sections which I find can make it more evident.

So my summary would be that while it involves some change in established habits and possibly a config change or two I would expect that most people could adjust to producing .wav or .flac files as for previous dramatic readings and that this would give the highest quality end result, but that if anyone can not handle these formats then, fingers crossed, using .mp3 files might introduce only limited loss of quality if the number of edits can be kept low.

What I'd suggest is that contributors try exporting to .Flac or .Wav instead of .mp3 and see if that works for them and only fall back to .mp3 if some insurmountable problem arises ?

All the best,

Padraig
Laurie Anne

chocoholic
LibriVox Admin Team
Posts: 13607
Joined: January 16th, 2007, 9:23 am
Contact:

Post by chocoholic » August 21st, 2008, 4:57 pm

Padraig, you make good points. I will say, though, that the uploader's capacity is not unlimited, and a few plays' worth of .wav files could overload it. As for .flac, I don't have that option on my version of Audacity (yes yes, I could update) and I have been hearing that there are other people who don't have easy access with their software either. Perhaps we should investigate .flac further, but we do need to keep things simple for people who don't have much experience with recording (or in some cases, with computers). Not to squash your idea; just saying this is a good subject for discussion. My personal opinion: I want to make things easy for everyone who handles the files (including prooflisteners) and I am not too concerned about a little bit of quality degradation (which I have not really noticed in completed plays anyway), so I prefer to use mp3. If there are individual files that need more work than simple volume adjustment (say, a challenging background noise), then we could ask for those files to be re-exported in .wav format.
Laurie Anne

icyjumbo

Post by icyjumbo » August 22nd, 2008, 2:50 am

I'd like to throw in a data point here. I'm afraid it's going to be approximate, as I don't remember exact facts and figures, but I do remember the lesson, which is that a noise-cleaned mono FLAC is about twice the size of a 128kbps MP3.

I record in full stereo WAV, even though I have only one microphone. It's just that my recording device will ONLY record stereo.

I recorded a test piece, which I saved as FLAC, and it was about half the size of the original WAV.

In Audacity, I split the stereo tracks, deleted the right channel, and then marked the remaining track as mono.

I then saved the track, just as it came out of the recorder, as FLAC. It was about half the size of the first FLAC, so about a quarter of the size of the original WAV.

Then I noise-cleaned and saved in FLAC again. This file was about 20% the size of the original WAV. It seems a good bet to assume that noise in the file will make the compressed version (i.e. FLAC) bigger than a quiet file. In MP3, I suspect that it will degrade the sound quality rather than increasing the size, as the compression-rate will remain constant.

Finally, a 128kbps MP3 is widely observed to be about 10% of the size of the original WAV.

From this I suggest that a noise-cleaned mono FLAC should be about twice the size of a similar 128kbps MP3, but won't suffer any degradation. As a result of this experiment, I changed my process. I now noise-clean as the first thing I do, then save the file as FLAC before any editing. Once the project is finished, I delete the intermediate files and save the final version in FLAC too. I therefore keep the original and submitted versions in full resolution in as small a format as reasonably practicable.

I mean this solely as a data point in the discussion, not an argument for one point of view or the other.

And YMMV, of course :)

PaulW
Posts: 2274
Joined: February 24th, 2007, 5:00 am
Location: Austin, Texas

Post by PaulW » August 22nd, 2008, 7:00 am

chocoholic wrote:[snip] (yes yes, I could update) [snip]
Audacity does allow you to have multiple versions on you system at the same time. You could continue to use your current version for recording and editing, and the newer version just for noise cleaning and conversion to/from the .flac format.
Paul
[b]DPL: [url=http://librivox.org/forum/viewtopic.php?t=12852]Brigands of the Moon[/url]; [url=http://librivox.org/forum/viewtopic.php?t=13151]Brain Twister[/url][/b]

chocoholic
LibriVox Admin Team
Posts: 13607
Joined: January 16th, 2007, 9:23 am
Contact:

Post by chocoholic » August 22nd, 2008, 1:00 pm

PaulW wrote: Audacity does allow you to have multiple versions on you system at the same time. You could continue to use your current version for recording and editing, and the newer version just for noise cleaning and conversion to/from the .flac format.
True enough... I use the Beta that way for NC already. Never noticed it has .flac capability since I rarely use it.
Laurie Anne

PaulW
Posts: 2274
Joined: February 24th, 2007, 5:00 am
Location: Austin, Texas

Post by PaulW » August 22nd, 2008, 2:04 pm

chocoholic wrote:True enough... I use the Beta that way for NC already. Never noticed it has .flac capability since I rarely use it.
Yep. When you export, select FLAC Files in the drop-down file type listing in the save box, and you can set options by clicking the options button. When I did a test on one of my own wav files, I only noticed a couple of K-bytes difference between the default level of 5 and the "best" level of 8. But that was with a raw, un-noise-cleaned file.
Paul
[b]DPL: [url=http://librivox.org/forum/viewtopic.php?t=12852]Brigands of the Moon[/url]; [url=http://librivox.org/forum/viewtopic.php?t=13151]Brain Twister[/url][/b]

textus
Posts: 4
Joined: May 19th, 2008, 2:34 am

Post by textus » September 2nd, 2008, 4:24 am

this is a perfect example of an old axiom, which is that any lossy encoding (such as MP3) should be performed from an uncompressed / lossless source

converting from MP3 -> WAV does not affect the quality, but essentially what is happening is that the same audio is being repeatedly compressed with the MP3 algorithm, and unfortunately each pass will add additional artifacts

it may not be practical at the moment to use FLAC for all recordings, but in the interests of future planning and additional compression options, it would be wise to move in this direction

Post Reply