Listening Test using Radiohead Clip

Notes by ff123

This clip comes from the opening chords following the introductory "noise" of Radiohead's "Anyone Can Play Guitar." I have placed it on my Audio Samples Page as "cymbals.flac". The file has been losslessly compressed using the FLAC format. It has lots of cymbals and snare drums, along with a sustained guitar chord. It appears to be very tough on the various codecs, even causing the FhG Alternate and mp3enc codecs to produce dropouts in sound at the 128kbs setting (similar to the dropouts described by davel at caffeine.co.uk when mp3enc encodes the "blackbird" clip; see the Lame Quality and Listening Test Information page). The worst performer at 128kbs is QDesign and the best is FhG Alternate-bad (as implemented in MusicMatch 5). In comparison, the Alternate version is among the worst sounding.

At 160kbs, I start to practice a technique which seems to make it easier for me to identify the encoded signal: I add in about 5 seconds of silence between successive auditions of X. I don't know if there's any reason why this should work, but continuous listening seems to "fatigue" my ear. After a period of silence, I perceive the artifacting at the beginning of the clip more distinctly. I thought rapid switching would actually be more effective. The "Alternate-bad" codec in MusicMatch 5 performs the best at 160kbs -- I can't distinguish it from the original.

At 192kbs, all codecs are operating pretty much at full bandwidth (see signatures page), which is as it should be, because artifacting is hardly noticeable at this bitrate. Even QDesign sounds good. Interesting is Xing, which I tried twice, on two different sittings. I scored 11 of 16 on both ABX tests. This could indicate that Xing is starting to sound better than the others at high bitrates, but the differences are so small between the codecs at this point that it's hard to be sure.

As I test the last encoder ("Alternate-bad") at 192kbs, I notice a clicking noise at the beginning of the clip. The clicking is there on the original too, but it's softer. On the basis of the loudness of the click alone, I am able to identify the coded signal 16 of 16 times. I go back to "Alternate" and am now able to pick out 16 of 16 on that clip as well, just by listening for the louder click. Previously, I had identified only 13 of 16 with Alternate. I determine that the clicking noise has always been there whenever I play any sound clip from any program; it's just that I've never paid attention to it before. I'm not sure why the click is louder for the encoded files, but now that I know this little trick, there's no way I could do a fair ABX test unless I do something about it. So I decide to insert 0.5 seconds of silence at the beginning of every clip, post-encoding. Thus, the clicking will sound the same for every file, and it will not mix in with the sounds from the clip itself.

Later, I discovered that the click is only audible after I have played a sound. That is, if I play a clip of silence, then the following sound I play will not have a click. Very weird.

I decide to re-run all of the 192kbs tests. The new runs are indicated in red. I most cases, when I rerun the test, the number of correct responses goes down. Interesting is my third run on Xing, in which I can now hear a watery sound, and am able to pick it out 15 of 16 times.

At 256kbs, the difference between the codecs are negligible to my ears. That doesn't mean that real differences don't exist -- just that I can't hear them.

Return to ff123's home page