I'd like to introduce a medium preset, for medium bitrates (and medium quality). Here are the switches I'm currently playing with:
--alt-preset standard --lowpass 17.5 -b 80 -Y --athaa-sensitivity -11 --nsmsfix 3 -V3
disclaimer: i KNOW that this won't be transparent, it's not the purpose
I'd like your opinions about this.
edit: following options were tested but gave bad results:
--interch 0.17
I'm all for it. I want to be able to put more songs on a CD than I can right now with --alt-preset standard, without being forced to use ABR or devise my own (undoubtedly crappy) commandline.
Because, let's just face it, ABR can be almost as much as a waste or bottleneck for some samples as CBR.
Thanks for posting that line, Gabe. I'm going to give it a try.
Yeah, I with NeoRenegade. I tend to put all my albums from a particular group on one CD. I find that most of the time there are around 8-10 albums to do, sometimes they all fit on with aps, sometimes I've got 50 megs to much. So an 'apm' for the just too much case would be great, and saves me from cooking up my own (undoubtedly crappy) commandline aswell.
Why not just use --alt-preset standard -Y
That knocked about 40MB off of one Metallica CD compared to APS.
Originally posted by indybrett
Why not just use --alt-preset standard -Y
That knocked about about 40MB off of one Metallica CD compared to APS.
Agreed, I remember reading a reply by Dibrom to a post (question) about using alternative switches in conjunction with --aps in order to save bits. He explained that -Y effectively lowpasses at 16KHz and only encodes higher freqs when it wouldn't require too many bits (I don't remember much more than that). But I do recall Dibrom saying that decreasing the lowpass is the most graceful way to retain quality and reduce bitrates.
Gabriel- What kind of bitrates are you thinking of? If a new --ap switch is implemented for the 150~190Kbps, it might as well just replace --r3mix
edit- grammar
Well, I am interested in working on a lower bitrate preset at some point (still probably going to be awhile before I'm done with anything), but I want to make sure that it is the best way to do it and that it is thoroughly tested.
Out of curiosity Gabriel, how much of a lower bitrate does that line give than just --aps -Y as someone already asked? Can you give some figures?
Also, I think part of the problem with that line is that you tune many of the settings just beyond where a lot of the internal --alt-preset tunings will no longer kick in at all. For example, the --athaa-sensitivity and --nsmsfix (and even the -V to some extent) settings you suggest are effectively well out of range of the "trigger" area so I'm doubting they will have much effect at all anymore..
Those switches seems to provide encoding in the range of 130-180kbps
Yes, you're right some switches are pushed to their sensitivity limit (msfix and athadjust)
Honestly the change in msfix doesn't affect the bitrate that much.
But for the ath adjust, it's another thing. In some tracks ap standard uses about 30 kbps for this adjustment.
This is quite high, and for something in the range of 130-180 it seems to me that this is way too high. A -11 change is high, and so the sensitivity is quite low, but it's still there. I think that a minimum sensitivity is better than no ath adjustment.
For the internal settings: yes, if you mean thanging the quantization on high ath adjustment levels or high ms values, probably it won't appear. Probably the only internal thing left which really does something is the different quantization on short and long.
Originally posted by Gabriel
Honestly the change in msfix doesn't affect the bitrate that much.
If that is so, then I would seriously suggest leaving the default settings that --aps uses. They were tuned that way for a reason, because they specifically eliminated artifacts in some samples, and if the bitrate savings is not significant (more than 10kbps on average), it would seem better to leave it as it is I think.
But for the ath adjust, it's another thing. In some tracks ap standard uses about 30 kbps for this adjustment.
This is another matter... and you're probably right that there isn't going to be a way to get the bitrate down significantly without doing something related to the ath or athadjust...
From my experiments with --alt-preset normal (which produced some incredibly low bitrates and still sounded good on quite a bit of music), it's actually possible to get away with a significantly raised ATH (if I remember, it was around --athlower -18 or so.. I'll have to look it up again) on most music such as pop or rock. The problem comes on quieter music where this high ATH causes noise pumping and ringing problems. I think this could actually be solved if the athadjust code was modified in a manner to try and be much more aggressive in these cases and only for this particular preset...
The only other issue, which is sort of related, was that the preset was more likely to have problems with -Z in some cases, but again mostly only on the lower volume music, so in those cases it had to be switched off... this worked but not as well as I would have liked.
The --alt-preset normal was actually on the right track I think... with a properly tuned athadjust, it could probably work great even nearly as good as --aps, at around the bitrate you are speaking of..
For the internal settings: yes, if you mean thanging the quantization on high ath adjustment levels or high ms values, probably it won't appear. Probably the only internal thing left which really does something is the different quantization on short and long.
Yeah, this is what I'm talking about. It really wouldn't be so difficult to find new thresholds for the new preset, it would just take some time to test it all.
I don't really like the idea of highering the ath (btw 18dB seems huge). I think that if you higher the ath, then you're relying on athadjust to lower the ath to its original value. It will lower the ath quicker than normally because the sound level is relatively closer to the ath than in normal situation.
So you have a slight delay before ath is restored to its original value. After it's restored (ie lowered by athadjust), athadjust as a more limited range to lower again.
So it seems to me that highering the ath is similar the lowering the athadjust, except that there will be a little delay.
So it seems to me that directly lowering the athadjust is a better way.
Btw the bit graph of encspot gives you a good visual idea of the athadjust effect.
Originally posted by Gabriel
(btw 18dB seems huge)
Looking back through some of the information, it was actually 16db. And yes, it was a pretty huge jump, it still worked pretty well in a lot of cases (coupled with much more usage of -X3).
So it seems to me that directly lowering the athadjust is a better way.
This could be... I'll have to run some experiments again, it's been so long that I don't remember all of the results I found messing with the athadjust stuff. Probably won't be till next weekend though. At any rate, if you are seeing a 30kbps savings from this and it doesn't harm quality very significantly on quieter music, then coupled with -Y, it could be a good solution.
Btw the bit graph of encspot gives you a good visual idea of the athadjust effect.
Yeah.. Those graphs would have been really handy back when I was working on revamping the --alt-presets
each time preset std is the first parameter
test.wav
std: 176.2
lowpass 17.5: 176.2
Y: 176.1 (ok, there is nearly no high freq content in this track)
athaa-sensitivity -11: 158.2
nsmsfix 3: 165.5
-V3: 163.2
cumulated switches: 136.7
cumulated -b 80: 136.2
r3mix: 140.4
-------------------------------------------------
big.wav:
std: 205.2
lowpass 17.5: 203.4
Y: 203.3
athaa-sensitivity -11: 201.0
nsmsfix 3: 203.9
-V3: 192.2
cumulated switches: 184.8
cumulated -b 80: 184.6
r3mix: 174.4
--------------------------------------------------
fatboy.wav
std: 265.3
lowpass 17.5: 264.0
Y: 264.8
athaa-sensitivity -11: 263.1
nsmsfix 3: 260.8
V3: 256.0
cumulated: 247.5
cumulated -b 80: 247.0
r3mix: 209.2
-----------------------------------------------------
t1.wav
std: 167.4
lowpass 17.5: 166.6
Y: 165.9
athaa-sensitivity -11: 153.6
nsmsfix 3: 166.3
V3: 158.0
cumulated: 144.3
cumulated -b 80: 140.5
r3mix: 138.0
------------------------------------------------------
spahm.wav
std: 289.4
lowpass 17.5: 288.2
Y: 285.7
athaa-sensitivity -11: 284.2
nsmsfix 3: 286.8
V3: 285.7
cumulated: 276.3
cumulated -b 80: 275.1
r3mix: 190.3
--------------------------------------------------------
stair.wav (a small part of Stairway to heaven - a guitar riffle)
std: 219.4
lowpass 17.5: 210.0
Y: 189.0
athaa-sensitivity -11: 218.1
nsmsfix 3: 218.6
V3: 199.0
cumulated: 173.8
cumulated -b 80: 173.8
r3mix: 215
-------------------------------------------------------
peaceful:
std: 215.1
lowpass 17.5: 212.0
Y: 209.7
athaa-sensitivity -11: 209.3
nsmsfix 3: 215.0
V3: 200.4
cumulated: 192.5
cumulated -b 80: 192.5
r3mix: 187.2
--------------------------------------------------------
velvet:
std: 227.2
lowpass 17.5: 218.6
Y: 218.3
athaa-sensitivity -11: 226.8
nsmsfix 3: 226.9
V3: 213.9
cumulated: 200.7
cumulated -b80: 201.7 ???
r3mix: 197.0
Hello,
i just tested my standard samples in and it averaged ~151 kbit
But the quality suffers to much in my opinion
Most degrading sounds this --interch 0.17 Try it with peaceful
for example.
What does this switch try to do? I never heard about it before.
regards
Wombat
Originally posted by Wombat
Most degrading sounds this --interch 0.17 Try it with peaceful
for example.
What does this switch try to do? I never heard about it before.
It's Takehiro's new switch, inter-channel masking (only in 3.92alphas iirc).
http://www.geocrawler.com/archives/3/7323/2002/2/0/7860123/ (http://www.geocrawler.com/archives/3/7323/2002/2/0/7860123/)
http://www.geocrawler.com/archives/3/7323/2002/2/0/7860199/ (http://www.geocrawler.com/archives/3/7323/2002/2/0/7860199/)
"This enables LAME to calculate the inter-channel masking which means - The masking in left channel sound made by sound in right channel and vise-verse".
I think Takehiro meant this switch to be used with low bitrate (under 128kbps), because the channel separation gets worse, but quality at low bitrate may become better.
Thanks John,
just like you said, used with aps here it simply doesn´t seem to work at all
and even isn´t meant to work here.
I will play around with it.
Wombat
You're right about the interch options. Probably the target bitrate for normal is too high to choose this compromise.
Let's edit the upper results without interch...
....and assemble all the results in 1 post
I have the feeling that -Y mainly reduces on electric guitar samples. Anyone found a non-guitar sample where it provides a significant bitrate reduction?
I also posted the r3mix bitrates as a comparison. After all, replacing r3mix is perhaps possible, as it seems to be in the same bitrate range.
Gabriel,
I saw your bitrate reductions using all of those switches. Did you happen to check the reduction in sound quality associated with your custom commandlines? Which ones gave the best trade-off of bits saved vs quality loss? Also some of us over at r3mix discussed the name of "medium" here:
http://66.96.216.160/cgi-bin/YaBB.pl?board...&num=1019269361 (http://66.96.216.160/cgi-bin/YaBB.pl?board=general&action=display&num=1019269361)
RD.
Originally posted by Gabriel
I have the feeling that -Y mainly reduces on electric guitar samples. Anyone found a non-guitar sample where it provides a significant bitrate reduction?
I also posted the r3mix bitrates as a comparison. After all, replacing r3mix is perhaps possible, as it seems to be in the same bitrate range.
I have in the past. The samples I've seen which benefit from -Y in regards to bitrate savings seem to vary pretty wildly. I wouldn't limit it to electric guitar samples, but instead to samples where high frequencies may be prominent and where the signal is noisy which, not surprisingly, is usually what leads to bloated bitrates in the first place. I've found samples from various electronic and experimental genres, ambient music, even classical music with high amounts of background noise, in addition to rock/metal.
I can't provide any samples off the top of my head this very moment, but maybe I can soon. I just got done moving again and so now I can finally get back to working on LAME some (including working on this lower bitrate preset and performing some listening tests), so I'll see if I can give some more clear examples of this.
If you'r searching for a song where -Y gives a HUGE bitrate reduction, I have a song (Ark/Burn the sun/Torn [Metal prog]) where it takes down from 234 to about 175 (!!!) kbps with aps.
If you can't find it I can find a way to upload it to someone.
?
How to remove a post?
Originally posted by Gabriel
a little up...
At this point, I'm waiting to see what happens with nspsytune2. I've been talking to Naoki some and I'm supposed to be doing some help with the tunings but I'm still waiting for him to email me a build with his latest enhancements. I don't think it makes sense to work seperately from this because the possible improvements could have a fairly significant impact and are really the more appropriate ways to try and improve quality (which would lead to similar quality at lower bitrates).
up
What about 3.93 and --preset medium?
I know that Dibrom would like to use some of Takehiro's changes for preset medium. Those changes should be in 3.94, not 3.93.
Do you think that it would be a good or bad thing to introduce a preliminary --preset medium in 3.93? (personnaly, I'd go for it)
I am asking because I hope that with the 3.93 release, it will be the end of our --alt-preset vs --preset strangeness, and it should be the beginning of a clear path for most people: use --preset
sounds like a plan to me. B)
Considering "--preset"
Will we see: (?)
--preset ABR xxx
--preset CBR xxx
--preset standard
--preset medium
--preset extreme
--preset insane
will standard, medium & extreme be possible to combine with --vbr-mtrh?
^ Personally I'd like mtrh to be default for all modes, it seems many use other codecs due to 'LAME's slowness'
-Y (default for "medium" and optional for rest?)
In 3.93 --alt-preset is the same as --preset
Apologies for this layman question, but can you give us an 'advance preview' of what switches -preset medium will be equivalent too, to use now?
I've found that -alt-preset standard -V5 -b96 gives me around the filesize I want, but after reading this thread (most of which was over my head), it seems that innocuous playing with switches might not be so innocuous, quality-wise...
v
The only thing that is sure is that over standard, it will includes -Y and a lower -b
Can you comment on whether -aps -V5 -b96 is a questionable choice for the time being? I'm just wondering whether that introduces artifacts, or simply acts to make irrelevant other desired switches contained in 'standard'. I.e. would I be better off changing V5 to Y for now?
You can consider -Y and -b96 as safe.
But probably not -V5.
What average bitrate are you targetting, and what bitrate do you have when using aps on your music?
I'm targetting 128-160 range, ideally, both for portability and sharing when appropriate. aps is significantly above that (180-200? can't recall offhand).
Well, even with -Y -b96 you will probably be quite above (around 160-190)
I would suggest you to use --alt-preset 150 in the meantime
Hm, yeah, before, I was using 144 -b96, but what I was missing was a 'global' look at the file, where quiet, sparse sections could sacrifice the resolution for the busy, higher-freq sections. (I encode lots of dynamic music - jazz, experimental, etc - which at this point you may be saying, "well, then, settle for a higher bitrate!").
But, if there's no perfect answer for what I'm asking for, ok, c'est la vie.
So ABR at 144 (or whatever) is probably better than aps -V5, even though the latter often reaches up into higher freqs than the former (judging by the encoding histogram)? Is this because V5 screws up other stuff?
Mainly because aps was not designed to be used with V5
I was surprised to see some posts in this thread where the Y switch hardly changed the file from aps. On my files, it is about a 25% reduction in file size (From 200-220 aps down to 150-180). I cannot tell the difference in quality between the 2 on my Ipod. Has anyone else experienced a difference in quality? On what equipment?
I would love a true VBR with a switch that can crunch the file size down to 150-160 with similar quality to the Y switch. ABR is not a good substitute for VBR. I can tell the difference b/t ABR 160 and aps -Y.
Can anyone really tell the difference between aps and ape?
I was surprised to see some posts in this thread where the Y switch hardly changed the file from aps. On my files, it is about a 25% reduction in file size (From 200-220 aps down to 150-180). I cannot tell the difference in quality between the 2 on my Ipod. Has anyone else experienced a difference in quality? On what equipment?
I would love a true VBR with a switch that can crunch the file size down to 150-160 with similar quality to the Y switch. ABR is not a good substitute for VBR. I can tell the difference b/t ABR 160 and aps -Y.
Can anyone really tell the difference between aps and ape?
The size difference will depend on the high frequency content of the original wav. If there are lots of HF, the output size difference will be bigger.
Being able to tell the difference between -Y and withuot -Y depends on your ear's capabilities to hear HF as well as your playback equipment's capability to play those back.
Given that MP3 is a format generally played on less-than-stellar equipment - PC speakers, portables, car stereos, etc. - you don't necessarily need perfect transparency, not that MP3 is capable of delivering that anyway. In this light, throwing out signals above 16KHz isn't tragic. Sure, some might notice the omission of these very high frequencies in a direct A-B comparison but since MP3 wasn't really designed to handle 16KHz+ signals (no scalefactor for sfb21), save the bits and toss the signals.
Because I don't use MP3 for anything serious, I always use the -Y switch with the presets. It's a good compromise.
The problem then is: Downloading VBR-files doesn't mean I have good quality. For CBR files I can see the difference but not for VBR files.
Maybe you can add a small description to every file sourcecode (extreme, standard, medium, portable..) which can be read by programs like EncSpot ect. Otherwise I can't be sure wherther I have a portable mp3 file or a extreme...
can someone tell me what -preset medium ended up as?
can someone tell me what -preset medium ended up as?
Sure; it ended up being --preset medium .
Now being serious,
medium is a new preset you can find in versions 3.92; 3.93.1 and the future 3.94, based in VBR modes, not ABR. It produces an average bitrate arround 165 - 170kbps; so, it's not transparent as --preset standard or --preset extreme can be; but IMHO it sounds great, and is the way to go for portable devices.
If your question was about what command line arguments are the equivalent of this preset; I think the answer is
none. The presets are theaked operation modes that cannot been reproduced with a command line, that's why they sound better
Why not use -m j instead of --nsmsfix?
-m j will reduce filesize more than --nsmsfix; --nsmsfix maybe isn't optimal at the bitrates used.
And with -m j you could increase the lowpass to 17.7/18 Khz, increasing the quality.
Why not use -m j instead of --nsmsfix?
-m j will reduce filesize more than --nsmsfix; --nsmsfix maybe isn't optimal at the bitrates used.
And with -m j you could increase the lowpass to 17.7/18 Khz, increasing the quality.
Hmm, it sounds like you are confused.
-m j basically means "use joint stereo". It doesn't have anything to do with reducing filesize (by itself...of course joint-stereo is more efficient than straight stereo).
--nsmsfix tweaks the mid/side stereo switching and accepts a decimal value from 0 to "infinity". It's not a toggle. Values closer to 0 make LAME use more L/R frames as well as modifying some masking algorithm. The default --nsmsfix value for LAME is 3.5. --alt-preset standard uses 1.38, I believe. If this sounds confusing then just ignore this switch.
I know that --nsmsfix it's a setting that use less L/R frames than -m j.
I said to use -m j instead of --nsmsfix, because I think that -m j (default Joint Stereo, which use more L/R frames) can help to decrease filesize more than --nsmsfix.
On a sample I've obtained an average bitrate of 150kbps with --nsmfix and of 145kbps with -m j.
Then I've said that with the save of bits obtained with -m j we can set an higher lowpass, increasing the quality.
I've found that by using 3.90.2-msvc and "--alt-preset fast standard -Y" I get files between 155 and 180 roughly. The "fast" brings down the siza a bit and so does the msvc compile. Never been happier:-)
IMO -Y degrade the sound, you should try another ways to decrease bitrate.
previously
-alt-preset standard
now
--preset standard
Why don't use
-standard
isn't it easier this way
IMO -Y degrade the sound, you should try another ways to decrease bitrate.
The -Y switch might decrease the sound quality (I for sure can't detect any sound loss) but it's still one of the best ways to do it. IMO
I know that --nsmsfix it's a setting that use less L/R frames than -m j.
I said to use -m j instead of --nsmsfix, because I think that -m j (default Joint Stereo, which use more L/R frames) can help to decrease filesize more than --nsmsfix.
On a sample I've obtained an average bitrate of 150kbps with --nsmfix and of 145kbps with -m j.
Then I've said that with the save of bits obtained with -m j we can set an higher lowpass, increasing the quality.
I still don't understand what you are saying.
All of the alt-presets and presets (except for insane) use -m j joint stereo by default. You do not need to add the -m j switch.
--nsmsfix is used alongside -m j; neither is used instead of the other. For instance, --alt-preset standard uses "-m j --nsmsfix 1.38". The first switch says use joint stereo, the second says set the switching threshold to 1.38.
If you are creating your own commandline (while using psytune) and do not specify --nsmsfix, LAME defaults to --nsmsfix 3.5. Remember, this is only for 3.94 alphas, not 3.90.2.
So you could modify --preset standard with "--nsmfix [some value larger than 2]" and this will reduce the bitrate (but I'm not necessarily recommending this).
Ok.
Just 1 question:
What treshold I should use with --nsmsfix to obtain the same amount of L/R Frames of -m j ? (-m j used without any other switch)
Ok.
Just 1 question:
What treshold I should use with --nsmsfix to obtain the same amount of L/R Frames of -m j ? (-m j used without any other switch)
The default --nsmsfix value is 3.5 (unless you are using one of the named presets).
Therefore, "-m j" and "-m j --nsmsfix 3.5" are the same.
So I've got a question that perhaps I should post elsewhere. I got down off my high horse of ogg and started using mp3, largely b/c a friend gave me a portable that reads mp3's off of cd-r. My optimum quality to size ration for such listening seems to be fine at around -alt-preset cbr 160 (depending on the difficulty of the music for encoding). -aps is bigger than I feel like going. But, given the bitrate at which I'm working, is there a good vbr setting that's actually intelligent vbr (I'm using cbr b/c I don't see abr as giving much marginal benefit). How is the -r3mix setting? I've never used it.
rip waves w/ EAC
encode using Lame 3.93 dll w/ CDex as frontend
Trim any beginning silence, and ending silence if live or supposed-to-be-gapless transition with mp3trim
mp3gain @ 89 dB
Tag using Tag
My optimum quality to size ration for such listening seems to be fine at around -alt-preset cbr 160
My suggestion is to use "-preset 160" (ABR) instead of "-preset cbr 160". ABR provides much better quality than CBR. Or try "-preset medium". I think it will render files around 160 kbps or so. It is (of course) a VBR preset and much better than CBR 160.
--alt-preset standard -Y only slightly degrades the sound compared to --alt-preset standard. Why not try it.
That will save plenty of bits with only a little reduction in high-end frequency response and is a "safe" method regarding artifacts, so it's a big bang for your buck.
If lowering bitrate is that important, it's probably the safest method (using --lowpass at around 16 kHz is also safe regarding artifacts)
As mentioned, medium is worth a try too if you're using a newer lame version, and this already effectively uses -Y to save bits, plus some more tricks to save a few more without losing too much more quality, and ABR at 160 kbps (--preset 160) is also best if the bitrate consistency is the main factor for you.
All these remove the very highest audible frequencies from all or most of the music using -Y or --lowpass. This provides more bits to avoid audible distortion and other artifacts in the most important frequencies where MP3 is most efficient, and is still better than good FM radio.
It seemed to me that --preset medium delivers the slowest encoding I have ever experienced with LAME.
I know it's not a problem for most people, because of P-IVs, Athlons etc, but not so with me
But if using the preset really gives the best possible quality at bitrates around 150-160, I would nevertheless use it.
Immo
This shouldn't really be attached to this thread, but...
I second DickD's oppinion... use:
--alt-preset standard -Y
This will still preserve SOME high frequency sounds (unlike --lowpass 16) and massively decrease the bitrate. You may also want to consider using --alt-preset fast standard -Y beacause it encodes way faster, and tends to be a few kb smaller. Especially if you use dibrom's MSVC binary instead of the ICS version (saves a few more kb)
if these are STILL too large, i highly recommend ABR over CBR. It sounds MUCH better around 128-160kbps.
It should be noted that using the "fast" standard is going to not provide as high a level of quality as the normal version. Same thing goes for the MSVC compile vs the ICL one. Whether or not you will notice this is a different matter, but it should at least be made clear that these drops in bitrate are not "free".
Dibrom, I knew that was the case for fast standard and -Y, but are the differences between the 2 compiles audibly different? Like, have people (with better ears than I) been able to ABX --alt-preset standard between the two compiles? Just curious, I use the ICL one for its speed, but i'd be happy to justify the added bitrates to myself
I have tried:
--preset medium
in Lame 3.93.1 (from Mitiok site), and it behaves differently than I expected. Looks like the -b80 is not included there anymore (a few 64 bitrate frames are there). I made mp3 with the long command:
--alt-preset standard --lowpass 17.5 -b 80 -Y --athaa-sensitivity -11 --nsmsfix 3 -V3
The files are different; this long command brings files slightly smaller, of course, without frames smaller than 80.
Is this normal, or a bug in 3.93.1 implementation?
Is --preset medium going to be further modified in the 3.94 release or can we expect it to stay the same (or almost the same) as in the current alpha? In other words, has the medium (or medium1 which was it finally?) preset been settled?
BTW, is there an estimate on the release of 3.94?