Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Suggest ~160kbps VBR setting. (Read 45419 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Suggest ~160kbps VBR setting.

Hi,

Looking at the suggested settings/presets I am dismayed at the lack of a VBR setting that operates at a lower bit-rate.

There is APS standard, which on my music easily averages over 200Kbps. In fact my latest Rush albums are averaging over 220Kbps.

I can see noisy metal taking over 220kbps, but a song like Rush-Nobody's Hero taking 230Kbps?

For me, it calls into question the validity of the VBR model if the answer to lower than 230KBps encoding is to use the ABR/CBR settings.

ABR doesn't seem much better than CBR after watching it encode 90+ % of its samples at the target rate.

Surely a VBR setting can be cooked up that can encode the above track at ~160kbps in VBR and still sound better than  160 ABR.

PS - I want to use this setting in my compact Flash based portable, in my car etc. My max space average is 160Kbps do to the size of the portable memory. But I want the best quality in that space.

Peter

Suggest ~160kbps VBR setting.

Reply #1
Take a look at this thread for some info on how to lower bitrates for --alt-preset standard. Basically, it suggests using a lower lowpass to trim the filesize.

How much lower should the lowpass be? Well, I used the tonesweep from
ff123's site (a great resource, IMO). Look for the "test your hearing cutoff" pages. I spent almost half an hour trying to ABX his mustang clip at 15KHz and couldn't do it. Using his tone sweep, I found my hearing cutoff is around 15.5KHz (unless I really turn up the volume). So for me, I lowpassed about ten albums (noisy rock) at 16KHz and got ~180kbps with APS, a savings of about 30-40kbps compared to a full-range APS, and I couldn't hear a difference.

Dibrom has mentioned that he is working on lower-bitrate VBR for Lame, as well as a more "flexible" ABR. I'm really looking forward to these new developments, but I have no idea when he'll be done with them.

Suggest ~160kbps VBR setting.

Reply #2
You could also try the --r3mix switch, which results in smaller file sizes while still maintaining pretty good quality for VBR MP3's. Yes, I'm aware that some of you will argue that --alt-preset standard is "much better," but on the majority of music (outside of test samples), the averager listener may be hard-pressed to notice a difference.

Suggest ~160kbps VBR setting.

Reply #3
Quote
Originally posted by Negative Zero
You could also try the --r3mix switch, which results in smaller file sizes while still maintaining pretty good quality for VBR MP3's. Yes, I'm aware that some of you will argue that --alt-preset standard is "much better," but on the majority of music (outside of test samples), the averager listener may be hard-pressed to notice a difference.
Well, that depends. If a Joe average listener plans to listen using headphones and is even slightly sensitive to pre-echo, then one should use better setting than --r3mix or better format than MP3..
Besides --r3mix can hit even higher average bitrates than APS with metal etc. music. You certainly are not gonna get about 160kbps average with --r3mix when encoding metal.

Because of the deficiencies in MP3-format (missing scalefactor for SFB21) and not perfect psychoacoustics, decent quality VBR mode hitting 160kbps average for metal, is pretty much impossible at the moment. CBR/ABR modes don't suffer from the SFB21 problem in that regard at all.
Juha Laaksonheimo

Suggest ~160kbps VBR setting.

Reply #4
Why are there still so many --r3mix followers? It should be considered a deprecated switch and you shouldn't use it anymore because there are better alternatives: the alt-presets or a custom switch list for your personal needs.

Although I am committed to Musepack/MPC, if I wanted to make some good MP3s, I'd use "--alt-preset extreme -Y". Very high quality at a rather reasonable bitrate.

Suggest ~160kbps VBR setting.

Reply #5
Quote
Originally posted by mithrandir

Although I am committed to Musepack/MPC, if I wanted to make some good MP3s, I'd use "--alt-preset extreme -Y". Very high quality at a rather reasonable bitrate.



Define reasonable bit rate. My player has very limited storage. I want an average of 160kbps or less.

I realize VBR will not turn in an exact average bit rate and will vary with material, but is it not unreasonable to have a VBR setting lower than APS standard for compression purposes.

I am seeking a setting that will encode the above mentioned track (nobody's hero) at 160kbps instead of 230kbps. I would think it would sound better than ABR/CBR 160kbps. If not then I question the validity of the VBR models.

Peter

Suggest ~160kbps VBR setting.

Reply #6
I say alt-preset extreme -Y creates files with a reasonable bitrate because it's comparable to --r3mix and aps, i.e. somewhere around 192kbps. Most people will agree that you need at least 192kbps on average to get near-transparent quality from MP3.

If you need 160kbps, then use --alt-preset 160. I don't know why you want to dwell on VBR. ABR uses variable framesizes, so it won't "waste" bits or underallocate on some frames like CBR. ABR makes a whole lot of sense if you have a final bitrate in mind.

From what I understand, the lower bitrate VBR modes simply use a less-sensitive ATH curve. That's probably not the best way of lowering bitrate. When you use ABR, you are telling the encoder: give me the best quality possible with this allocation of bits. When you use VBR, you are telling the encoder: here's the ATH curve I want to use, apply it to the processing of each frame.

Suggest ~160kbps VBR setting.

Reply #7
Quote
Originally posted by guidryp
I am seeking a setting that will encode the above mentioned track (nobody's hero) at 160kbps instead of 230kbps. I would think it would sound better than ABR/CBR 160kbps. If not then I question the validity of the VBR models.
Well, I don't know the song, but any such Lame vbr-mode would need to use -Y swithch. Then you would lose pretty much the over 16kHz high freq content. I'm not sure if even --r3mix -Y gives you that bitrate with that track, and if you start to tweak --r3mix for lower quality, it will probably fail pretty fast pretty ugly at least for many other tracks... That all depends on your hearing though. But if the comparison is to ABR-160kbps..

VBR setting which would give average 160kbps with metal would probably fall to low very often with different types of tracks or even with quieter sections, because the quality and bitrate is heavily dictated by the masking threshold from the psychoacoustic model (which is far from perfect).

ABR mode uses different method for determining the bits needed - ABR-160 basically first takes 160kbps for a base bitrate and then uses so called Perceptual Entropy (PE) algorithm which gives an rough estimate how difficult the frame about to be encoded is. PE then decides whether the frame is difficult or easier, but because PE is pretty rough estimate, the adjustment doesnt happen very often, only when it's pretty much needed to raise the bitrate it will, and only when it's pretty much safe to use lower bitrate, it will. Of course CBR/ABR modes also use the masking threshold of psychoacoustic but it does not dictate the bitrate like in VBR mode. Masking threshold in CBR/ABR mode will only tell the encoder where the bits should be placed, while in VBR-mode it dictates in addition the resolution (amount of bits -> bitrate) which should be used.

So if we compare abr-160 and vbr-mode which is tweaked to give 160kbps average with metal, the vbr-mode will more often use too low bitrate frames, because vbr's bitare is mostly dictated by the non-perfect psychoacoustics and you have given permission to the encoder to base the bitrate on higher masking threshold (lower resolution) in order to achieve your desired vbr average bitrate. Then again this kind of vbr-mode sometimes might quality wise correctly give higher frames when needed eg. for transients, but the problem is the bitrate falling too low from time to time. Anyway, the absolute requirement for this kind of Lame vbr setting is the use of -Y switch.
Juha Laaksonheimo

Suggest ~160kbps VBR setting.

Reply #8
Quote
Originally posted by mithrandir

If you need 160kbps, then use --alt-preset 160. I don't know why you want to dwell on VBR. ABR uses variable framesizes, so it won't "waste" bits or underallocate on some frames like CBR. ABR makes a whole lot of sense if you have a final bitrate in mind.
.


I guess I am more representative of the average listener than the majority in these forums. My goal is not necessarily transparency buy non-offensive compression. I want the best size/quality tradeoff with the emphasis on size.

We have a pool of MP3's at work, mine are ripped with APS-standard. Some are from the net with 128 CBR, the rest are a friends DLL based lame VBR setting of some sort that yields about 140 kbps on average.

I set a giant playlist and let em rip randomly. The 128 CBRs grate me even when I am not paying attention. The VBR'd 140kbps are fine 90% of the time, 10% have easily noticeable artifacts. My APS standards are perfect. But average about 210Kbps.

I would like something in between the 140 kbps and 210 kbps, but closer to 140 kbps, since I also want to use them in a portable with limited space. The best use of bits which gives something that gives good casual listening. 

If I can tolerate the non optimised ~140 kbps lame dll based encoding 90% of the time. I figured a ~160 kbps tweaked/optimised exe version should move it up to 98% or better.

I would prefer VBR, since in theory it has the potential to offer the best bang for the bit. It still seems wastefull to me to use CBR or very similar ABR settings.

-Y is a real possibility. While I can hear 20k on frequency sweep, I don't know if I can hear the difference of a 16k cutoff in music.

I will have to wait until my good computer (fast with spdif out) returns before I do some testing.

Peter

Suggest ~160kbps VBR setting.

Reply #9
Quote
Originally posted by JohnV
Well, I don't know the song, but any such Lame vbr-mode would need to use -Y swithch. 

VBR setting which would give average 160kbps with metal would probably fall to low very often with different types of tracks or even with quieter sections, because the quality and bitrate is heavily dictated by the masking threshold from the psychoacoustic model (which is far from perfect). 


I was suprised at the particular song using 230Kbps since it is mostly acoustic guitar and vocals. I don't consider it metal.

The -Y switch is something I will try. I should make it clear that I would like something that has a smaller output size than --r3mix.

I guess I want APS Sub-Standard.

The problem I have with ABR, is that it looks like a marketers attempt at VBR. It varies frames a bit, but it appears like it is not enough to be significant.

Philosophically VBR should deliver the best bang for the bit, so I get stuck on that. Even a low tweaked VBR will still use a number of 320kbps frames that I don't see from ABR 160. When watching them each encode. So it seems that VBR is spending more heavily on the difficult sections. I guess it would be annoying to me that ABR 160 would still use 160 kbps to record silence.

Apparently sometime in the future there may be a lower VBR preset as well as more variable ABRs, either of which might make me happier.

But what to use in the interim?

Something with -Y -b112


Tin eared Peter

Suggest ~160kbps VBR setting.

Reply #10
Quote
Originally posted by guidryp
The problem I have with ABR, is that it looks like a marketers attempt at VBR. It varies frames a bit, but it appears like it is not enough to be significant.

Philosophically VBR should deliver the best bang for the bit, so I get stuck on that. Even a low tweaked VBR will still use a number of 320kbps frames that I don't see from ABR 160. When watching them each encode. So it seems that VBR is spending more heavily on the difficult sections. I guess it would be annoying to me that ABR 160 would still use 160 kbps to record silence.

You appear to more concerned with what you see on the console output rather than what you hear. It is not acceptable, for instance, to say that one encode is better than another because it has twice as many 320kbps frames. If both files are the same size, then the one with more 320kbps frames is making compromises in other frames. Which is better? You really can't say unless you rate each file in a blind test.

You say ABR appears not to vary frames enough to be significant, but that is just an unproven hypothesis unless you have real listening test results that validate it. You would benefit from some blind listening tests so that you can remove any hint of subjectivity and pre-determined notions from your decision-making. I'd like to make two WAV files of Nobody's Hero: one decoded from a 160kbps ABR MP3 and another decoded from a 160kbps VBR MP3. You wouldn't know which is which and would have to say what you thought of each. THAT'S where you want to take this: forget about what you "feel" about ABR, the proof is in the pudding. If you think a 160kbps VBR file sounds better than a 160kbps ABR in such a test, then that's the encoder method for you in this case. But until you perform such a test, you shouldn't claim ABR is somehow flawed for various gut-feeling reasons while people here are telling you objective reasons why VBR can falter at lower bitrates and why ABR is the better choice.

Suggest ~160kbps VBR setting.

Reply #11
Yes, VBR would use more bits for harder sections, but that kind of Lame vbr mode which would give average 160kbps with metal is going to fall too low pretty often and will introduce clear distortion, especially if -Y (ignoring of SFB21 noise shaping) or very high --ns-sfb21 value (high masking for SFB21)  is not used. If -Y is not used, you would have to compensate the high amount of bits given to all SFBs because of the over 16khz content (due to SFB21 lack of scalefactor) by increasing the overall masking threshold (lowering overall resolution with higher -v switch) significantly, which would lead to overall lower quality. So there's no way to get good high freq response and low average bitrate with MP3 VBR. This is a problem in MP3 technical details (lack of scalefactor for SFB21), and it can't be fixed.

Of course if there was an easy way to create a working decent quality lowbitrate vbr-mode, it would have been done and  commonly in use. But sadly this is not the case at the moment. Don't know if Dibrom has some tricks in his pocket..?

Also with higher bitrate VBR it's not always only difficult sections that get 320 or 256kbps frames especially with metal. Metal music can have quite high masking tendency for low/mid bitrates, but because lots of high frequencies may have to be coded, the overall bitrate will skyrcoket again because of SFB21 issue. It directly means that simply because of loud over 16kHz content, lower frequency content may have to be coded with much higher resolution than would be otherwise necessary, it's grazy, but blame the missing scalefactor for SFB21...
Juha Laaksonheimo

Suggest ~160kbps VBR setting.

Reply #12
Quote
Originally posted by mithrandir

make two WAV files of Nobody's Hero: one decoded from a 160kbps ABR MP3 and another decoded from a 160kbps VBR MP3. 

But until you perform such a test, you shouldn't claim ABR is somehow flawed for various gut-feeling reasons while people here are telling you objective reasons why VBR can falter at lower bitrates and why ABR is the better choice.


Good points and I would do that, but I am looking for the optimally tweaked command line options that would give me my ~160Kbps VBR.  So far I have -Y.

AFAIK there was talk of an ABR vrs VBR test, but it still hasn't occurred.  Has there been an ABR vrs CBR test?

I still don't like the fact that ABR modes, like CBR modes will take still take the same amount of bit to encode very simple material when savings might have been gathered.

It also occurs to  me that if I have a VBR mode delivering 160 kpbs on average, and if it takes ~180 kbps to encode more difficult songs it should do a better job than ABR 160 on that material.

Are the models so bad at lower bit rates that a 180 kbps VBR encoding will still sound inferior to 160 kbps ABR? If this is the case I will certainly avoid VBR like the plague.

The VBR problem should be on simpler material where VBR gives 140 Kbps. It would then be inferior to ABR 160. I can see that, but would it still be acceptable, more or less objectionable than the more difficult material limited to 160kbps by ABR?

Given good models A VBR scheme should give a better quality/bit rate ratio than any locked bit bit rate scheme, when used on a range of material. 

What you are telling me is that the models are not that good at less than the alt-preset standard level (200Kbps+ average).  Ok I get it.

Peter

Suggest ~160kbps VBR setting.

Reply #13
Quote
Originally posted by JohnV
Yes, VBR would use more bits for harder sections, but that kind of Lame vbr mode which would give average 160kbps with metal is going to fall too low pretty often and will introduce clear distortion, especially if -Y (ignoring of SFB21 noise shaping) or very high --ns-sfb21 value (high masking for SFB21)  is not used.


Rush == Metal?? Where does Progressive Rock fit into this scheme (Classic Yes, Rush, Genesis, Jethro Tull)?

I don't really think I listen to anything termed metal. I have maybe 1 Iron Maiden album.  Is that what you mean by Metal? I hate anything with a wall of loud distortion.

Maybe my first test should be to see where -lowpass x makes a difference for me. If I can't hear the difference above 16KHz, there is no point encoding it.

Peter

Suggest ~160kbps VBR setting.

Reply #14
Quote
Originally posted by guidryp
Rush == Metal?? Where does Progressive Rock fit into this scheme (Classic Yes, Rush, Genesis, Jethro Tull)?

I don't really think I listen to anything termed metal. I have maybe 1 Iron Maiden album.  Is that what you mean by Metal? I hate anything with a wall of loud distortion.

Maybe my first test should be to see where -lowpass x makes a difference for me. If I can't hear the difference above 16KHz, there is no point encoding it.
I said I don't know this band. It's of course not the issue what type of music this is to people, it's more an issue what type of music it's to the psymodel.
Juha Laaksonheimo

Suggest ~160kbps VBR setting.

Reply #15
Well after a bit of testing, I am going with:

--alt-preset standard  -V3 -b96 --lowpass 16

So far the album I am ripping is averaging ~170kbps.

For me this seems just about transparent.  I went ff123 page and when I do the sweep I can hear right to the end(When I did sweep on my setup at work it stopped at 16khz). But testing on his mustange clips and every music track I have ripped, show it is impossible for me to distinguish --lowpass 16 (except on sweep).

I also ran the various test waves (castanets etc.) and it seemed to cope very well.

I am running spdif digital out from my Nforce MB to my denon 1802, an into to my Koss R80s Phones or Paradigm monitor 3s. Sounds perfect to me.

I was going to test the corresponding ABR setting, but I don't see the point anymore. Since I really cant distinguish this from the original WAV file (in music tracks). I did note a difference here or there on some specific test clips, but they were extremely minor.

Tin ears I guess. But thats alright with me.

Peter

Suggest ~160kbps VBR setting.

Reply #16
Quote
Originally posted by guidryp
Well after a bit of testing, I am going with:

--alt-preset standard  -V3 -b96 --lowpass 16

So far the album I am ripping is averaging ~170kbps. 


If you are trying to save bits and you are going to use --lowpass 16 anyway, you might as well use -Y instead.  Not only should it save you a few more bits than --lowpass 16 (there's no need to combine them either), but it should also still encode a little bit over 16khz sometimes... though not while increasing the bitrate.

Try this instead:

--alt-preset standard -V3 -b96 -Y

Suggest ~160kbps VBR setting.

Reply #17
For whatever reason, I remember --lowpass 16 using less bitrate than -Y. Adding -Y to --lowpass 16 used exactly the same amount of bits. My conclusion was that the lowpass superseded -Y.

I will try that again on a different track to see.

Suggest ~160kbps VBR setting.

Reply #18
Quote
Originally posted by guidryp
For whatever reason, I remember --lowpass 16 using less bitrate than -Y. Adding -Y to --lowpass 16 used exactly the same amount of bits. My conclusion was that the lowpass superseded -Y.
Seems that you didn't understand the SFB21 problem with vbr I was trying to explain. If you use only --lowpass 16, the lower frequencies may still bloat the bitrate because of the SFB21 vbr problem (there still remains few MDCT coefficients over 16kHz which can bloat the bitrate when doing the bit allocation/noise shaping).

-Y affects to the bit allocation of all frequencies (SFBs) by disabling over 16kHz noise shaping.

Here's the more exact explanation of the SFB21 vbr problem: http://www.hydrogenaudio.org/forums/showth...10724#post10724
Juha Laaksonheimo

Suggest ~160kbps VBR setting.

Reply #19
Quote
Originally posted by JohnV
Seems that you didn't understand the SFB21 problem with vbr I was trying to explain. If you use only --lowpass 16, the lower frequencies may still bloat the bitrate because of the SFB21 vbr problem (there still remains few MDCT coefficients over 16kHz which can bloat the bitrate).

-Y affects to the bit allocation of all frequencies (SFBs) by disabling over 16kHz noise shaping.


I had read the theory, but my test result did not hold to it. In theory, theory and practice are the same. In practice, theory and practice are different. 

I just repeated my test on a different track with the same result. Not dramatic but still has lowpass 16 doing better than -Y. Perhaps -V3 is affecting things in a way you weren't expecting?

Rush - Mission:  -alt-preset standard -V3 -b 96
No Lowpass or Y = 6787 bytes
-Y                    =  6393
--lowpass 16  =  6351
Both                =  6351

I am still concluding that there is insufficient energy with lowpass 16 to trigger the extra bits needed for proper rendering of lower frequencies. 

Peter

Suggest ~160kbps VBR setting.

Reply #20
Thought I might try this for myself.

Without going into specifics, APS -Y yielded a smaller file size than r3mix on my particular music sample.

So there ya go all you r3mix lovers. Now you've got your preset.

I'm tempted to use this myself, but I think I'll wait and see what Dibrom is cooking up before I go and encode a few thousand songs. I'd hate to have to do it all over again in a month 

As a side note. I encoded "Fatboy" as well. APS -Y came out the same size as APS. R3mix came out smaller. I expected this.

So, it looks like a good compromise. Smaller file size by losing some high frequencies that are possibly inaudible anyway, and still maintaining the quality in the audible range.
flac > schiit modi > schiit magni > hd650

Suggest ~160kbps VBR setting.

Reply #21
Quote
Originally posted by guidryp

I had read the theory, but my test result did not hold to it. In theory, theory and practice are the same. In practice, theory and practice are different. 

I just repeated my test on a different track with the same result. Not dramatic but still has lowpass 16 doing better than -Y. Perhaps -V3 is affecting things in a way you weren't expecting?
Peter
Hmm, I think -V3 may very well have an effect that the bloating doesn't happen so often. The reason for the bloating is relatively high resolution deman for SFB21 compared to other SFBs, which leads to resolution increase in all SFBs, but when you lower the overall resolution demand by increasing -V it may very well help make it happen more rarely.

Still I think if you test even more tracks, you'll find a track that bloats clearly with --lowpass 16 but doesn't at -Y.
If you don't care about over 16kHz then for best results you should use both.  If you wan't to retain even somekind of over 16kHz freq response, you should use only -Y which prevents file size bloating.
Juha Laaksonheimo

Suggest ~160kbps VBR setting.

Reply #22
Just my opinion, but if you're gonna do an abrupt cutoff at 16KHz, make sure the MP3's are for your ears only.  Please, don't share (ever!).

Suggest ~160kbps VBR setting.

Reply #23
Quote
Originally posted by fewtch
Just my opinion, but if you're gonna do an abrupt cutoff at 16KHz, make sure the MP3's are for your ears only.  Please, don't share (ever!).


Share? Most of my friends encode at 128 Kbps and don't notice anything wrong with that. They think I am crazy for using 180Kbps to encode a song.

The crap of the net is equally as bad or worse. I hear terrible 192Kbps encodes, with dropouts etc...

Can you consistently ABX a 16KHz cutoff in music? I can't, and I can hear out past 20KHz. This is sitting home with digital out into my reciever and decent Phones and speakers. Critically trying to spot the difference.

I am definitely getting the impression that this is the "golden ear" forum, with the kind of people who can hear the difference in speaker cables.

Thankfully I am not that highly tuned. I enjoy the sound  coming through my speakers with plain old spool 16 gauge speaker wire. :-)

Peter

Suggest ~160kbps VBR setting.

Reply #24
Reading the suggestions of Dibrom and others regarding the use of -Y along with aps brings me to a question.

Many of the mp3's I encode are from FM radio sources, and FM of course cuts off around 15khz. That being the case, would Dibrom's suggested setting of --aps -V3 -b96 -Y be optimal for encoding FM sourced WAVs?

Scott