Amongst the podcasting fraternity the debate over “what is the best MP3 encoder to use for podcasts” has raged for a quite while, although the general consensus seems to have swung in one particular direction. Is this a fair situation? Can YOU actually tell the difference yourself?
The de facto standard for audio podcasts is the ubiquitous MP3. MP3s are a convenient way of storing digital audio data in a highly efficient manner. They do this via a method of lossy compression using “perceptual coding” and “psychoacoustic” techniques to discard certain audio data that the human auditory system cannot hear.The result saves data and can sound virtually indistinguishable from the original source material to the majority of listeners.
A useful analogy to help explain this is if you imagine a rock band playing at full volume with a triangle player lurking amongst the amplifier cabinets. You are very unlikely to hear the triangle! In this scenario the MP3 encoder decides that the tour bus is a bit full and fires the triangle player, in so doing saving much needed space.
A typical MP3 encoded at 128 kbps uses approximately one tenth of the data to store audio compared to uncompressed CD specification source material. This makes MP3s very attractive if you’re a podcaster as you can retain reasonable audio quality whilst reducing the download times of your podcasts and their associated storage requirements.
MP3 Encoder Choices for Podcasts
As a podcaster you essentially have two choices of which MP3 encoder to use to encode your podcasts; either the Fraunhofer encoder or the LAME encoder.
Fraunhofer – The Original MP3 Encoder
The Fraunhofer Society in Germany were amongst a group of organizations in the Motion Picture Experts Group (MPEG) that developed the MP3 (MPEG-1 Audio Layer III) encoding format in the early 1990s. Fraunhofer IIS own and control the world-wide licensing for the MP3 format. So, contrary to popular belief, the MP3 format is not an “open” or “free” format.
The Fraunhofer MP3 encoder is not available for download or purchase as a standalone encoder. It is usually only found embedded in third-party software (such as iTunes and Windows Media Player) with the applicable licenses paid by the software manufacturers. However, as the patents for the key MP3 encoding technologies were filed back in the early 1990s it won’t be too long before these expire and the format becomes patent free.
LAME – A Copycat MP3 Encoder
LAME is a free software codec (compression engine) for encoding MP3 audio files. The name LAME stands for “LAME Ain’t an MP3 Encoder”, which is a recursive name (insomuch as the name refers to itself – a naming convention popularized by the geeky fraternity).
Although LAME is considered free, it is actually only released via source code as an “educational description of an MP3 encoder” in an effort to circumnavigate the fact that it implements certain patented technologies of various organizations, most notably the Fraunhofer Society. Therefore any software that uses the LAME source code as an MP3 encoder falls under the applicable patents and may require licensing.
Fraunhofer vs. LAME
So, why the choice; don’t all MP3 encoders sound the same?
Unfortunately no. Not all encoders were created equal and indeed even different versions of the same encoder can sound different!
The wisdom of the crowd (if indeed crowds possess collective wisdom) is that the Fraunhofer encoder is best at encoding constant bit rate (CBR) MP3s whilst the LAME encoder is better at encoding variable bit rate (VBR) MP3s.
By better we mean that for a given bit rate (or quality setting in the case of variable bit rates) one encoder will be “perceived” to sound better, or closer to the original audio source material with less digital artifacts.
Podcast MP3 Requirements
To retain compatibility with older playback equipment, podcast MP3s are usually encoded using constant bit rates. The “majority” of podcasts are also usually encoded in mono at 64 kbps..
As podcasts are most commonly encoded using constant bit rates this has led to the podcasting community opting for the Fraunhofer MP3 encoder as their encoder of choice.
Judge for Yourself
So, are you a sheep and quite happy to follow the crowd and accept the prevailing wisdom, or would you like to decide for yourself?
You would? Excellent! Then have a listen to the following and choose which MP3 version you think sounds the closest to the original.
Original Source
This source recording was recorded at CD quality 44.1 kHz sample rate, linear PCM 16-bit depth with no processing of any fashion (e.g. compression, EQ etc.):
Or, download: CODEC.wav [0′ 06”, 557kB]
Alternative MP3 Encodings
Three MP3 encodings are presented below, one encoded by the Fraunhofer encoder and two by the LAME encoder.
Why two from the LAME encoder?
Well, the vast majority of podcasters who produce LAME encoded MP3 files are likely to do so using the LAME encoder via the Audacity audio editing program. During the MP3 encoding process Audacity passes a set of commands to the LAME encoder, one of which is a quality setting (q3 in this case).
An alternative way of producing a LAME encoded MP3 is by using the WinLAME front end for the LAME encoder. WinLAME on its highest quality setting for a constant bit rate MP3 uses the q2 setting (better than the default Audacity setting).
As most Fraunhofer / LAME encoder podcast comparisons have probably been using the Audacity LAME q3 encoder setting, it might be useful to add the WinLAME q2 encoder setting as an additional comparison.
The LAME encoder version used for the production of the LAME based MP3s below was v3.98r for both the Audacity and the WinLAME versions.
The command strings passed to the LAME encoder for the test files below were:
- -m m -q 3 -lowpass 16.5 -b 64 [Audacity]
- -m m -q 2 -lowpass 16.5 -b 64 [WinLAME]
-m m = stereo mode; mono
-lowpass 16.5 = low pass filter set at 16.5kHz
-b 64 = bit rate of 64kbps
-q 2 or -q 3 = algorithm quality selection
If you want to examine media files such as MP3s to see what encoder was used for example, a handy little utility to do just this is called MediaInfo.
In no particular order:
[powerpress url=”https://www.richardfarrar.com/audio/CODEC-1.mp3″]Or, download: CODEC-1.mp3 [0′ 06”, 51kB]
[powerpress url=”https://www.richardfarrar.com/audio/CODEC-2.mp3″]Or, download: CODEC-2.mp3 [0′ 06”, 51kB]
[powerpress url=”https://www.richardfarrar.com/audio/CODEC-3.mp3″]Or, download: CODEC-3.mp3 [0′ 06”, 51kB]
Now you’ve listened to all three, it’s time to make your choice. Which MP3 sounds best?
Once you have submitted your answer below, you will get to see the results of the poll so far.
Loading ...To find out which MP3 is which, click here.
In this short sample, I cannot tell the difference, at least on the crappy ear buds I keep here at work. However, I can tell you that over the course of a longer file with much wider frequency variation, I have received very noticeable differences between exporting to MP3 with Audacity vs iTunes and I will always use iTunes, at least until I work with different software.
It’s also worth noting that you will get different results listening on different headphones and speakers and even in playing from different players.
Hi Mat, You’re right that these are very short files and longer files with different source material would be easier to tell. Harpsichord music for example is very revealing of CODECs, but most podcasts are spoken word I would guess and probably very few contain any harpsichords 🙂
I can hear a little of a difference in the third one. My vote was on that for being Fraunhoffer.
What microphone did you use?
I may be remembering this incorrectly, but I thought the -q# switch only affected VBR and does nothing to CBR.
Hi Daniel,
As to whether you guessed correctly, there’s a link at the bottom of the article that will show you what the 3 CODECs were. I must confess I don’t hear any striking differences myself. May be different source material would reveal more significant differences, but the test was aimed specifically at the spoken word typical of most podcasts. Perhaps my ears are on the wrong side of 40 and a little worse for wear playing in rock bands in my youth 🙂
The microphone I used was a Rode NT1-A straight into a Focusrite Saffire firewire interface.
According to LAME’s official documentation (http://lame.cvs.sourceforge.net/viewvc/lame/lame/USAGE) the -q# affects the algorithm quality of the LAME encoder, so I would think it is worth experimenting with. The -V# is used to set the VBR quality:
I couldn’t tell the difference after many listenings of the files. I’m using Sony 7506 studio monitors, went to film school, and have been working with sound design and audio software for years, so to me I take it as evidence that the vast majority of listeners will discern no difference whatsoever.
Thanks for the article!
Thanks Allen. Listening on my pair of Beyer-Dynamic DT 770 Pro headphones I’m not really sure I can reliably tell the difference, so likewise I would guess the majority of people listening to podcasts on cheap ear buds are unlikely to notice a significant difference either.
I listened first using the built-in speakers on the 12″ Macbook (which are surprisingly good for crappy laptop speakers). I could not tell the difference. Then I tried my (crappy) Apple earbuds…and one was significantly better than the other two. I’m on the wrong side of 40 (heck, the wrong side of 50) but I didn’t go to rock concerts and don’t even play much music, so maybe that helped. Up until about 3 years ago I could still hear that annoying sound they created to bother the children.
I really enjoyed reading this and learning about it. Thanks!
Hi Allison, thanks for your feedback.
I seem to remember reading somewhere that females have a slightly extended frequency response to males of the same age, so this should be working in your favour too 🙂
Hallo Richard!
First thanks a lot for your great comparison!
With good headphones, it is possible to hear the difference between the LAME and Fraunhofer versions:
both LAME versions include noticeable noise and distortions during speech periods.
However, which exact LAME version and parameters did you use?
Here is your example file, CBR encoded with LAME 3.99.5:
https://auphonic.com/media/audio-examples/CODEC-4.mp3
(exact command: lame -h –cbr -b 64 CODEC.wav CODEC-4.mp3)
IMHO this sounds much better than your LAME examples (much less distortion). One could also say that it sounds better than the Fraunhofer example: “was used to encode this file” (4sec) contains quite some musical noise in the Fraunhofter version, which is not present in the LAME version (but there will always be some kind of artifacts at such low bitrates and it also depends very much on the content).
I know it is too late to re-do the comparison, but it would be great if you could include the file CODEC-4.mp3 to your blog post (feel free to just copy the file to your server). Maybe we could repeat such an experiment with more files.
We (at Auphonic) often get requests to include the Fraunhofer encoder, but I did not hear one file where the Franhofer CBR encoder was noticeable better so far – I would be happy if someone could send me such an example 😉
Greetings from Austria,
LG
Georg
Hi Georg,
Thanks for your comment and the alternative encoding. I used the LAME 3.98r encoder with the following settings:
-m m -V 4 -q 3 -lowpass 16.5 -b 64 [Audacity] CODEC-1.mp3
-m m -V 4 -q 2 -lowpass 16.5 -b 64 [WinLAME] CODEC-2.mp3?
I shall definitely have to upgrade my encoder to the newer version and give it a good check out. Judging by some of the comments on the Google+ podcasting communities, I think this debate is likely to rage on!
Richard