I've been lurking around here for quite some time now (about 18 months) and finally I have thought of a question.

A little while ago I bumped into someone on another board who assured me that both fatboy and castanets were considered 'solved'. Which I found interesting, because I decisively disagreed. Have I been misinformed, or are they indeed meant to be transparent?

I had already ABXed them to make sure, but the results are slightly incomplete now, because of -Z becoming recommended.

3.90.2

fatboy,
aps...14/15
ape...12/12
api...12/12

castanets,
aps...12/12
ape...12/12
api...12/12

In both cases, I found ape the easiest to ABX, taking literally less than 30 seconds to 12/12 (haha 'speed abx'), and api was certainly not audibly superior to aps, but I didn't try to ABX between the presets.

If they rely on -Z to improve them, then I'm awfully sorry for having obsolete results. I did 12/12 fatboy on aps -Z, but didn't try any others because I was just playing with switches (-Z wasn't recommended back then).
The computer I have at the moment has a disgustingly noisy line-out, so I don't think ABXing is a good idea for me at the moment.

Hopefully this isn't all old news... I haven't noticed these samples being mentioned for a while...

Lossy encoders won't be 100% transparent on 100% of music.

See this thread:

samples where aps or ape has a problem

This thread gave rise to the -Z switch, so many of these samples aren't problems anymore.



No one will guarantee that lame is transparent with any settings. But, overall, --alt-preset standard is the best you can do with mp3, unless you want to use --alt preset insane. Most people find it completely transparent most of the time.

There are rare samples where it isn't transparent, and individuals who find more of these samples than others. But it's a million times better than what most people find acceptable.


Check that you're ABXing properly - in particular, be careful how you decode your mp3s. Also, check that you're accounting for any delay, and that there's no clipping.

There's always MusePack and lossless.

Cheers,
David.

Lame has still small pre-echo problems with samples of this kind, so it's not strange that you can ABX them if you are sensitive to pre-echo. IIRC the -Z switch doesn't address those problems, but others, so they are still there.

The effect of the -Z switch is best heard with the fatboy sample. It souns better with it. But those are problem samples that exploit the weaknesses of the mp3 format, that's why they are abxable even at 320kbps.

Ok, cool. I thought it would be be some sort of limitation of the format or something, but just making sure.

I'm not sure if I'm hearing pre-echo or not. For castanets, I don't listen to the castanets, just the guitar chord at the start. It sounds like a different guitar, or a different recording of the same guitar. And fatboy, again, sounds different in tone quality (more hf noise or something).


I just decoded back to .wav in razorlame. Do you mean make sure to listen for artifacts rather than a delay playing the file? And clipping, introduced by the encode? That would be a fault in itself...?

Encoding does not cause clipping, decoding does. So if you don't replaygain before you ABX, you could skew your results.

For this reason I use Foobar's ABX tool since it won't allow you to test a non RGed sample.