Notes by ff123
These are examples of artifacting in audio codecs, some very easy to hear, others more subtle. The intent is to familiarize the listener with many of the types of defects which various encoders make. Sometimes a specific term is used to describe an artifact, for example "pre-echo" or "noise-pumping." Once you learn what these things actually sound like, you may find yourself noticing them occurring in other places more often.
The ratings categories are somewhat subjective, and some people may find it difficult to hear something which I rate as "very obvious" while other may find it easy to hear something which I find to be very subtle, or even inaudible. In general, if I could easily hear the artifacting through my labtec (i.e., crummy) computer speakers, I classified it as "very obvious." If a sample is only subtly different for me from the original, I will perform an ABX test to prove with a confidence of at least 99% that I can actually hear the difference.
Some listening tips: you might be able to hear artifacting better with a good pair of headphones unless you've already got a good set of speakers. I personally find it difficult to hear some of these artifacts on cheap computer speakers. Listen in a quiet environment. You may have to adjust the volume up a bit on some samples.
For samples with obvious artifacting, I will typically try to make samples available in mp3 format for the most convenient distribution. If such samples were initially encoded with another format, such as WMA8 or mp3PRO or Ogg Vorbis, etc., in addition to making them available in the original encoded format, I will transcode to mp3 using the Lame --alt-preset insane preset (which yields mp3 at a constant bitrate of 320 kbit/s). The original file (or a link to it) will always be available as a reference, although I will typically also make an mp3 "pseudo-reference" file (320 kbit/s) available which should serve well enough as a reference in most cases of obvious artifacting.
For samples with subtle artifacting, I will make only the original encoded file and the original reference file available for direct comparison. This will mean you will have to download more data to hear these examples since the original reference file will be either a WAV file or losslessly compressed. Also, you may be required to get various decoders for both the encoded and the losslessly compressed reference files.
Summary
| Very Obvious | Obvious |
| applaud (warbling) castanets (transient smearing) BlackBird (lo-freq dropout) dr4 (roughness) mc_sich_short (stereo collapse) 41_30sec (chirping) |
wait (flanging, warped) daughter (knocking) left-dist (fluttering) |
| Original Encoded File | Transcoded Encoded File | Original Reference File | Pseudo Reference File |
| applaud_radium128.mp3, 131 kB | N/A | applaud.wav, 1.46 MB (link: Lame Quality and Listening Test Information) | applaud_insane.mp3, 333 kB |
| Comments: Read the comments on the Lame Quality and Listening Test Information page. This sample is from the end of the live version of the Eagles's Hotel California. The popular FhG/Radium codec (which is a hack of Opticom's Producer Pro 2.1) was used at a bitrate of 128 kbit/s to make the original encoded file. Throughout the file, but especially audible in the first several seconds is a severe form of the "underwater" sound, which is a warbly, hissing artifact. | |||
| Original Encoded File | Transcoded Encoded File | Original Reference File | Pseudo Reference File |
| castanets_xing128.mp3, 106 kB | N/A | castanets.wav, 1.17 MB (link: Lame Quality and Listening Test Information) | castanets_insane.mp3, 269 kB |
| Comments: Read the comments on the Lame Quality and Listening Test Information page. The Xing encoder within Audiocatalyst 2.1 was used at 128 kbit/s to make the original encoded file. Xing does not use short blocks, which means that its temporal resolution is very coarse compared with more advanced codecs. The castanet transients are smeared, so that they sound like "sh" sounds instead of nice, crisp attacks. Castanets.wav is the classic "pre-echo" test case, so called because the artifacting occurs before the transients happen. If you listen closely to the pseudo-reference mp3 file (320 kbit/s) you might be able to tell that even at this bitrate, there is a difference from the original. | |||
| Original Encoded File | Transcoded Encoded File | Original Reference File | Pseudo Reference File |
| BlackBird_mmjb128.mp3, 97 kB | N/A | BlackBird.wav, 1.07 MB (link: Lame Quality and Listening Test Information) | BlackBird_insane.mp3, 247 kB |
| Comments: Read the comments on the Lame Quality and Listening Test Information page. This sample was encoded with Music Match Jukebox "very high" setting at 128 kbit/s. See my page at http://ff123.net/dropouts.html for more information about this type of artifact. The sound seems to cut out momentarily during the sample. The first dropout occurs right after the first arpeggiated piano chord. | |||
| Original Encoded File | Transcoded Encoded File | Original Reference File | Pseudo Reference File |
| dr4_q0.ogg, 108 kB | dr4_oggq0_insane.mp3, 649 kB | dr4.rar, 1.55 MB (link: Vorbis sound test) | dr4_insane.mp3, 649 kB |
| Comments: This sample is from Dire Straits Your Latest Trick from the Brothers in Arms album. It was encoded with Ogg Vorbis RC3 at quality -q 0. The initial trumpet sound in particular, sounds rough or harsh. It's as if little burrs were sticking to it and catching in places (to use a visual analogy). | |||
| Original Encoded File | Transcoded Encoded File | Original Reference File | Pseudo Reference File |
| mc_sich_short_mmjb320.mp3, 345 kB | N/A | mc_sich_short.flac, 923 kB (link: my samples page) | mc_sich_short_insane.mp3, 347 kB |
| Comments: JohnV found this sample, which is "Sichia" (de Lucia) track 4, 2:00-2:35 CD: Passion Grace and Fire (McLaughlin, Di Meola, de Lucia) shortened from http://www.minidisc.org/mc_sich.wav. Music Match Jukebox 6.0 using the "normal" setting was used at 320 kbit/s. This version of MMJB (version 6.1 and later fixes the problem) had a bug which collapses the stereo separation, and which is detailed at my page: http://ff123.net/fastenc.html. | |||
| Original Encoded File | Transcoded Encoded File | Original Reference File | Pseudo Reference File |
| 41_30sec_scmpx128.mp3, 481 kB | N/A | 41_30sec.flac, 3.43 MB (link: my samples page) | 41_30sec_insane.mp3, 1.20 MB |
| Comments: Filburt's test sample, the first 30 seconds from Dave Matthews Band, "#41." This sample gives a variety of different encoders problems. In this case, using the mp3 encoder called SCMPX, there is a chirping sound which accompanies the cymbals. | |||
| Original Encoded File | Transcoded Encoded File | Original Reference File | Pseudo Reference File |
| wait_xing128.mp3, 336 kB | N/A | wait.flac, 2.25 MB (link: my samples page) | wait_insane.mp3, 844 kB |
| Comments: 21 seconds from the beginning of Sarah McLachlan's Wait. The Xing encoder within AudioCatalyst 2.1 was used at 128 kbit/s. The prominent instrument is the ride cymbals, which is something which gives Xing particular trouble. In this sample, they sound warped, as if I were listening to a large piece of sheet metal being moved back and forth. This might also be termed flanging. | |||
| Original Encoded File | Transcoded Encoded File | Original Reference File | Pseudo Reference File |
| daughter_mp3enc128.mp3, 186 kB | N/A | Daughter.pac, 1.25 MB (link: Speek's Listening test at 128 kbps) | daughter_insane.mp3, 465 kB |
| Comments: Read the comments on Speek's page. Fraunhofer's mp3enc version 3.1 was used at 128 kbit/s (quality = 9) to make the original encoded file. The worst artifacting occurs about 5 seconds into the file, where I hear defects which sound like "knocking on wood." | |||
| Original Encoded File | Transcoded Encoded File | Original Reference File | Pseudo Reference File |
| left-dist_q0.ogg, 100 kB | left-dist_q0_insane.mp3, 574 kB | left-dist.flac, 1.09 MB or left-dist.wav, 2.51 MB (link: http://noa.tm/vorbis/left-dist/) | left-dist_insane.mp3, 574 kB |
| Comments: Read Daniel Resare's comments on the Vorbis development mailing list. Ogg Vorbis RC3 at -q0 was used to make the original encoded file. The artifacting which stands out is the fluttering. Tones which should sound clear, rather sound unsteady, just like tape flutter. Also, the beginning of the guitar plucks and the sounds of the guitarist repositioning his hands are "metallic." At 10 seconds, the loud vibration present in the original is less pronounced in the encoded version. This sample is easier to hear problems with if you turn up the volume a bit. | |||