I'm using EAC 0.95b4 with lame 3.97b2.

Recently, I switched from cbr 192k to using one of the VBR settings. I was pleased to discover that if the VBR setting resulted in an mp3 with an average compression of above, say, 170kbs the result would sound at least as good as the cbr 192 file. This increase in efficiency was cause for great celebration.

Problems arose when I started ripping 60s soul, 50s country, and anything recorded before that. Even if I used the V0 vbr setting, the files would come out at ~140kbs. The resulting mp3s sound crunchy and overcompressed. I see it as a problem when entire genres of music sound bad at the highest quality setting.

Before I return to using a constant bit rate setting, I wanted to ask if anyone had any advice. For example, might I have better results with this kind of material with one of the earlier 3.98 releases? Thanks in advance for any comments or suggestions.

Got ABX results?

BTW: Bitrate means nothing. Bitrate does not equal quality.

Most songs from the 60's and 50's are likely to be in mono and 160kbps - 192kbps for mono sound is a overkill. LAME is clever enough to detect and to encode mono frames (even mastered from mono to stero) at a smaller bitrate cause it knows it is a waste of space to encode mono frames at 192kps.

In my experience higher bitrates (up to a point) produce results that more closely approximate the sound of the source .wav file. How much difference it makes and what the cutoff point of perceptible difference depends on the material. So I do think bitrate means something.

Honestly, though, I don't care what the bitrate is as long as it sounds good. I was happy at first that the file sizes were so small, thinking that I'd be able to fit a ton of extra cds on my iPod. But the crunchiness sounds bad with this kind of music. Overcompressed 80s dance music, say, I find less offensive.

But I'll do an ABX test tomorrow and post the results.

In the meantime, could anyone tell me if a different version of lame would compress files without a lot of high frequency information less aggressively?

Thanks.

try adding -Y to your commandline

Make sure the files aren't clipping during playback; this could cause the "crunchiness" you describe. Use MP3GAIN to reduce their volume a notch, especially if you're not using foobar2000 as your player. (tis a small hazard of mp3.)

please upload the original wav file somewhere that sounds crunchy when encoded with LAME vbr and comes out at ~140 kbit/s

Cruchy and compressed aren't typical mp3 artifacts.

Yep, it would be nice if people could try to reproduce your findings. Short samples (shorter than 30 seconds) would be excellent.

I'm pretty sure this is a purely psychological effect. You think you "know" that low bit rate music must sound bad so that's how you hear it. Dont worry it's not just you, no-one is immume from being deluded in this way, that's why we suggest ABX testing. Also remember that those old tracks were probably not very high quality recordings in the first place. It's very easy to listen to just the mp3 and say, yuk that sounds ordinary it must be because of the low bitrate, without properly comparing it to the original which may well sound a little ordinary too.

By far the most likely thing that's going on here is that those low bitrate tracks are mono (as others have already said). I see this all the time when I'm encoding old 60's stuff. I've got some albums that are mostly stereo with a just few mono tracks thrown in and I can mostly tell which ones are mono by a quick glance at the resultant mp3 bitrate. In some cases I'd never previously noticed that a particular track was mono, but in every case where I've followed it up and tested the original track it has turned out to be mono.

It makes sense right. Lame (and most other mp3 encoders) efficiently encode what's common on both channels so there is only one copy of the information stored and not two. That's the advantage of (the default) joint-stereo encoding, otherwise with mono material you'd just have two copies (L and R) of exactly the same thing, the second copy of course being totally useless.

For example say you encoded a mono recording using "stereo" instead of "joint stereo" and the overall bitrate was say 190kbps, then each channel L and R would only have an available bitrate of 95kbps. So if you didn't store that redundant second copy then you'd get exactly the same quality in only 95kbps. This is more or less what joint stereo automatically does for you.

Anyway just do the abx test, I'm sure it will sort it out for you.

Lots of reasons possible here. Mono files recorded on old gear lacking in high frequencies will cause low bitrates. Higher bitrates are simply not required.

But if it bugs you adequately, just add -b 192 to your command line or whatever pleases you.That way,Lame will use a minimum of 192 for each frame.

As far as your subjective comments un crunchiness and over compressed, these need abx testing.

post small wavepack or flac clips of the originals that show the flaws so they can be tested.

Most likely lame is just doing its job.

I did a single-blind ABX test with the help of friends and the results were *not* statistically significant. I took a 30 second .wav from a George Jones song and put it up against LAME's -V0 vbr. The mp3 came out at 147kb/s. My friends insisted the .wav sounded noticeably better, but they did worse on the ABX than I did. We all felt that the pedal steel gets a bit buried in the mp3, and that the vocals sound harsher and more distorted. But without visual cues they were tough to distinguish. I was surprised.

So, thanks for the comments. I'm happy to say that you're all correct (except for the -Y suggestion by the Duck Man from New Zealand -- who I think was playing with me).

I'm glad to be wrong about this as my iPod's close to full. But I'm a little alarmed about what else I may be wrong about. The middle east? U.S. foreign policy? Billy Joel?

Thanks, all, I see big changes ahead.

did you try the -Y switch? e.g. -V0 --VBR-NEW -Y

I didn't because the only reference I found to this command indicated that it was intended for metal, to make the resulting files smaller not bigger. So I thought you were joking.

And since I couldn't reliably tell the difference in the ABX test, I figured it wasn't worth worrying about.

But I'll give it a try out of curiosity, and let you know.

If you want a "less agressive" filtering, try -k switch. It disables all psychoacoustic stuff; encodings are therefore visually better with more high frequencies content. But it's not recommanded at all: audible quality should be worse. -Y makes the cutoff stronger (or more agressive).

no, the -k switch just removes the lowpass (but the psymodel even with the -k switch enabled, esp. at low bitrates, still refuses to encode much into sfb21 (16k and up))

J.M.

I just tried with this quiet piano sample and -V2 --vbr new setting.
There is a serious difference introduced by the -k switch: lowpass is less agressive [or maybe no lowpass at all¹] but there's still visual ringing. So I was wrong and you were right. Thanks for the correction

A much... better (hmm...) switch is --noath: no "ringing" anymore, but a lowpass is definitely present.
A perfect (hmmm!) commandline would be: -V2 --vbr-new --noath -k. No ringing, no lowpass: it's visually very close to the original. Note that I didn't listen to the final results. It probably doesn't sound bad, but there's a strong loss in efficiency:



¹ I realized a bit later that the original sample is already lowpassed at 20500 Khz

@guruboolez I somehow fail to see in this post where someone mentionned the visual resemblance of 2 audio files being important. Actually I fail to see why it would be important anywhere, including in this post. I'm surprised at your post.

The original poster asked for this:
Complaint about HF information is typically a problem with the lowpass. And this complaint is usually based on a visual experience (a frequency editor is showing an agressive reduction of HF), not on a listening one.

So I just tried to answered to the original question. If someone want LAME files with a less agressive HF encoding, then -k and --noath are good switch to achieve this goal (-Y is intended to be more agressive with HF, not less). And the easiest way to check this is to look on the resemblance of two files. Am I clear?

Yep. Clear as cold ice.

lame 3.97 should do fine with -V 2 through -V 0
lame 3.90-3.93 don't have the -Y switch forced at lower -V settings, you might want to give one of those versions a try (with and without --nspsytune) and see how it goes.

other options to bloat the bitrate:
--athlower 10
--noath
--ns-sfb21 -8 (might be 8, I forget which way to go to make it spend more bits on stuff over 16khz)

It might even be worth it to try encoding with ABR and shutting off the psymodel completely.

A_Man_Eating_Duck wrote:
>did you try the -Y switch? e.g. -V0 --VBR-NEW -Y

That's is exactly the opposite of what he is trying to achieve. Do you even know what -Y does?

...coming to an audiophile (audiofool) message board near you soon!

(We probably shouldn't joke about such things!)

Cheers,
David.