Notes by ff123
Listening Test by bAdDuDeX
In a discussion of the comparative sound quality of mp3 encoders at bitrates of 128 kbit/s, I remarked that based upon the limited amount of listening tests I had performed, I would give the nod to Lame as the overall best-sounding encoder. I received strong disagreement with this assessment by bAdDuDeX, who made the following comments:
I think LAME is one of the worst sounding MP3 encoders at 128kbit/s. I would even take Xing over it at that bitrate. I wouldn't consider using LAME at anything short of 256kbit/s... Maybe 192 with Joint Stereo, but that's pushing it.
[The artifacts sound] like a ringing or something. It's hard to describe. It happens with Blade (worse though) and even with FastEnc...
MP3Enc messed up on a few test samples (like spahm.wav) but LAME sounds much worse on all regular music I've listened with.
I've downloaded tons of MP3's encoded with LAME and it occurs very frequently. It seems to always occur on quieter parts in a song... I just randomly picked a clip out (it wasn't planned) and it had tons of it. Download it here:
My Audio Samples Page as unt.mp3
It's United - Kill Yourself For Business from a Metal Blade (record label) sampler CD I have. I hear tons of ringing on the drum part between 7-10 seconds and it's throughout the whole song (just more concentrated on the drum part). I used 3.87RH, just like you wanted me to... I used -bw 19999 with MP3Enc and -h -k -d with LAME. There's 2 reasons for not using the default cutoff. First, it's not fair when comparing other codecs because they have a 16kHz cutoff with spikes to 20kHz as well. Second is because I feel a 128kbit/s MP3 needs at least a 16kHz cutoff to be even close to acceptable (to me).
[And later...] OK, just tried it with the default cutoff and the ringing is still there, just less.
Here's the MPC version [using the -insane profile]. It's basically transparent, so you can consider that the WAV.
My Audio Samples Page as unt.mpc
On the other hand, I used to use MP3Enc a long time ago when I still used MP3, and the warbling wasn't very frequent at all... And it produces no ringing whatsoever.
[after listening to the file generated with 3.88 --nspsytune; see below to download] I downloaded the file from your site and it sounds much better. There's still a little ringing but it's not too bad. I hope they implement --nspsytune soon... I'm sure sick of downloading all these 192's with artifacts everywhere.
Listening Test by Hans Heijden
The ringing artifact is apparently something that is heard by only a very few people who have exceptional hearing. But bAdDuDeX's assessment of Lame has been seconded by Hans Heijden who writes as follows on the r3mix.net forum (original post is here). The little "squeaks" described by Hans are the same as the "ringing" described by bAdDuDeX.
The thing that bothered me with lame (since summer '99), are the little squeeks around 15 kHz (present in whatever setting more or less). Reading the various boards in time, it appears only few people hear that. Anyway, finally --nspsytune seems to cure that!
Hans later wrote about listening to bAdDuDeX's sample clip:
Yes, I also hear the lame-ringing, but only around the 8th and 20th second, when the drums are 'rolling'.
I will download the mp+ now though I wouldn't trust if re-encoding that to mp3 makes much sense.
A few days ago I did some experimenting in CoolEdit with 1400.wav. The ringing is so strong in the left channel that I think I see it in spectral view. Doesn't look as dramatic as it sounds though. [see my page "Listening test of opera with male voice" for my own test of 1400.wav]
[responding to my comment on Hans's terminology] Well, ringing, squeaks, I think we mean the same sound. To me ringing would be a tone excited by a pulse for example, like steep filters can cause.
[after encoding from the mp+ file; I have recreated the file using CVS version 3.88 alpha dated Jan 4, 2001 and stored it as unt_lame388.mp3 on my Audio Samples Page] Ok, tried 3.88 -h -k -d --nspsytune and what I heard is gone now. Hope Badudex tries also since he hears trouble everywhere in the unt track.
Didn't hear a difference by adding -q 1 and -Y to this. There is also be -X3 and --athlower to be tried... and the combination with --abr of course!
If you look at the track in spectral view, you'll see that -k removes the lowpass filter but lame still limits to 16 kHz with only occasional higher frequency bursts. Lowering of softening the default lowpass filter of lame to remove the 'squeaks' doesn't work, they seem initiated by lower frequencies.
Comments by Naoki Shibata, author of nspsytune (original post on here on the mp3 encoder list)
At high bitrate CBR (and possibly ABR?), high frequency sounds (>16KHz) aren't encoded well. VBR doesn't have this problem. Also, nspsytune does really bad job on fatboy and spahm. Of course, I'm continuously trying to solve these problems.
It is not defaulted because it's still under construction, and not yet widely tested.